Hebrew Multiword Expressions: Linguistic Properties, Lexical
Timothy Baldwin - MSE...1 Multiword Expressions Structure of Course a. Introduction b. Computational...
Transcript of Timothy Baldwin - MSE...1 Multiword Expressions Structure of Course a. Introduction b. Computational...
Multiword Expressions
Timothy Baldwin
CSSE, University of Melbourne
1 Multiword Expressions
Structure of Course
a. Introduction
b. Computational syntax
c. Extraction/identification
d. Computational semantics/interpretation
e. Decomposability/compositionality
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
2 INTRODUCTION
INTRODUCTION
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
3 INTRODUCTION
Introduction: Structure
• Definitions, properties of MWEs
• Computational challenges
• Component tasks
• MWEs in NLP applications
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
4 INTRODUCTION
But First ...
• Research background:
? Multiword Expression (MWE) Project (CSLI, NTT CS
Labs. and Cambridge University)
? Jointly funded by NSF and NTT CS Labs.
? Primary project aim is to investigate different means
for encoding a variety of MWEs in precision grammars
? Visit us online at: mwe.stanford.edu
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
5 INTRODUCTION
What are Multiword Expressions (MWEs)?
• Definition: A multiword expression (MWE) is:
a. decomposable into multiple simplex words
b. lexically, syntactically, semantically, pragmatically
and/or statistically idiosyncratic
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
6 INTRODUCTION
Some Examples
• San Francisco, ad hoc, by and large, Where Eagles
Dare, kick the bucket , part of speech, in step, the
Oakland Raiders, trip the light fantastic, telephone
box , call (someone) up, take a walk , do a number on
(someone), take (unfair) advantage (of), pull strings,
kindle excitement , fresh air , ....
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
7 INTRODUCTION
MWE or not MWE?
... there is no unified phenomenon to describe
but rather a complex of features that interact
in various, often untidy, ways and represent a
broad continuum between non-compositional (or
idiomatic) and compositional groups of words.
(Moon 1998)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
8 INTRODUCTION
MWE Markedness
MarkednessMWE
Lex Syn Sem Prag Stat
ad hominem � ? ? ? �
at first � � � � �
first aid � � � � ?
salt and pepper � � � � �
good morning � � � � �
cat’s cradle � � � � ?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
9 INTRODUCTION
Indicators of MWE-hood
• Institutionalisation/conventionalisation
• Lexicogrammatical fixedness: formal rigidity, preferred
lexical realisation, restrictions on aspect, mood, voice,
etc.
? lexicogrammatically fixed MWE: kick the bucket
? lexicogrammatically fixed non-MWE: look like
? lexicogrammatically non-fixed MWE: keep tabs on
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
10 INTRODUCTION
• Semantic/pragmatic non-compositionality: there is a
mismatch between the semantics/pragmatics of the
parts and the whole
? non-compositional MWE: kick the bucket
? compositional MWE: at first
• Syntactic irregularity:
? syntactically-irregular MWEs: all of a sudden, the be
all and end all of
? syntactically regular MWEs: kick the bucket , fly off
the handle
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
11 INTRODUCTION
• Non-identifiability: meaning cannot be predicted from
surface form
? idiom of decoding (non-identifiable): kick the bucket,
fly off the handle
? idiom of encoding (identifiable): wide awake, plain
truth
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
12 INTRODUCTION
• Situatedness: the expression is associated with a fixed
pragmatic point
? situated MWEs: good morning, all aboard
? non-situated MWEs: first off, to and fro
• Figuration: the expression encodes some metaphor,
metonymy, hyperbole, etc
? figurative expressions: bull market, beat around the
bush
? non-figurative expressions: first off, to and fro
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
13 INTRODUCTION
• Single-word paraphrasability: the expression has a single
word paraphrase
? paraphrasable MWEs: leave out = omit
? non-paraphrasable MWEs: look up
? paraphrasable non-MWEs: take off clothes = undress
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
14 INTRODUCTION
• Proverbiality: the expression is used “to describe—and
implicitly, to explain—a recurrent situation of particular
social interest ... in virtue of its resemblance or relation
to a scenario involving homely, concrete things and
relations” (Nunberg et al. 1994)
? informality: the expression is associated with more
informal or colloquial registers
? affect: the expression encodes a certain evaluation of
affective stance toward the thing it denotes
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
15 INTRODUCTION
• Prosody: the expression has a distinctive stress pattern
which diverges from the norm
? prosodically-marked MWE: soft spot
? prosodically-unmarked MWE: first aid, red herring
? prosodically-marked non-MWE: dental operation
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
16 INTRODUCTION
• Substitutability: MWEs characteristically stand in
opposition to anti-collocations, i.e. expressions derived
through synonym/word order substitution which occur
with markedly lower frequency than the base MWE (or
not at all):
? non-substitutable MWEs: many thanks (cf. *several
thanks, *many gratitudes)
? substitutable MWEs: salt and pepper vs. pepper and
salt
? non-substitutable non-MWEs: common platypus
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
17 INTRODUCTION
Lexicographic Concept of “Multiword”
• Rough and ready definition: a lexeme that crosses word
boundaries
• Complications with non-segmenting languages (Japanese,
Thai, ...) and languages without a pre-existing writing
system (Walpiri, Mohawk, ...)
• Also, in English: houseboat vs. house boat , trade off
vs. trade-off vs. tradeoff
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
18 INTRODUCTION
Exercise: Spot the MWEMarkedness
Expression MWE?Lex Syn Sem Prag Stat
library cardat arm’s lengthold treeforeign direct investmentthe sunat [nine] o’clockto go bushgive a demokick the bucketonce upon a timeat homein the meantimeto read Shakespeare
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
19 INTRODUCTION
MWEs vs. Collocations
• A collocation is an arbitrary and recurrent word
combination
• Collocations can be semantically-marked (e.g. dark
horse) but tend to be compositional (e.g. strong coffee)
• Collocations are generally contiguous (binary) word
sequences (more often than not Ns)
• Word order variation/flexibility effects generally ignored
in collocation research
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
20 INTRODUCTION
MWEs vs. Terms
• (Technical) term = a lexical unit consisting of one or
more words which represents a concept inside a domain
• Terminology research primarily interested in the
synchronic dynamics of terminology, term formation
and terminological variation
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
21 INTRODUCTION
Challenges Posed by MWEs
• Syntactic idiomaticity
• Semantic idiomaticity
• Pragmatic idiomaticity
• Statistical idiomaticity
• Flexibility
• Productivity
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
22 INTRODUCTION
Challenges Posed by MWEs
• Syntactic idiomaticity
• Semantic idiomaticity
• Pragmatic idiomaticity
• Statistical idiomaticity
• Flexibility
• Productivity
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
23 INTRODUCTION
Syntactic Idiomaticityby and large
???
conjP Adj
by and large
ad hoc
Adj
? ?
ad hoc
wine and dine
V[trans]
conjV[intrans] V[intrans]
wine and dine
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
24 INTRODUCTION
Challenges Posed by MWEs
• Syntactic idiomaticity
• Semantic idiomaticity
• Pragmatic idiomaticity
• Statistical idiomaticity
• Flexibility
• Productivity
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
25 INTRODUCTION
Semantic Idiomaticity
kick the bucket spill the beans
die’ reveal’(secret’)
kindle excitement
kindle’(excitement’)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
26 INTRODUCTION
Decomposability and Syntactic Flexibility
• Decomposability = degree to which the semantics of
an MWE can be ascribed to those of its parts
• Consider:
*the bucket was kicked by Kim
Strings were pulled to get Sandy the job.
The FBI kept closer tabs on Kim than they kept on Sandy.
... the considerable advantage that was taken of the situation
• The syntactic flexibility of an idiom can generally be
explained in terms of its decomposability
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
27 INTRODUCTION
Challenges Posed by MWEs
• Syntactic idiomaticity
• Semantic idiomaticity
• Pragmatic idiomaticity
• Statistical idiomaticity
• Flexibility
• Productivity
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
28 INTRODUCTION
Pragmatic idiomaticity
• The Wheel of Fortune factor — how to represent the
jumble of phrases stored in the mental lexicon?
• The Monty Python factor — mish-mash of language
fragments which evoke particular events/individuals/memories
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
29 INTRODUCTION
Challenges Posed by MWEs
• Syntactic idiomaticity
• Semantic idiomaticity
• Pragmatic idiomaticity
• Statistical idiomaticity
• Flexibility
• Productivity
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
30 INTRODUCTION
Statistical Idiomaticity
unblemished spotless flawless immaculate impeccable
eye – – – – +
gentleman – – ? – +
home ? + – + ?
lawn – – ? + –
memory – – + – ?
quality – – – – +
record + + + + +
reputation + – – + +
taste – – – – +Adapted from Cruse (1986)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
31 INTRODUCTION
Statistical Idiomaticity and Dialect
• The arbitrariness of some MWEs is brought out well in
dialect differences (e.g. OzE vs. AmE):
? phone box vs. phone booth
? mail man vs. post man
? no through road vs. not a through street
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
32 INTRODUCTION
Challenges Posed by MWEs
• Syntactic idiomaticity
• Semantic idiomaticity
• Pragmatic idiomaticity
• Statistical idiomaticity
• Flexibility
• Productivity
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
33 INTRODUCTION
Flexibility
let the cat out of the bag
the cat was letout of the bag
let the fluorescent catout of the polythene bag
the cat that was letout of the bag
let the cat outa the bag
the cat is outof the bag
let the cat outof the bag
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
34 INTRODUCTION
Mapping the Boundaries of Flexibility
• Cline between full flexibility and full rigidity, e.g.:
Can/could you tell?
Are you able to tell?
*They might/ought to tell.
How might you tell?
*How ought they to tell?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
35 INTRODUCTION
Variation in Flexibility
• There is considerable variation in syntactic flexibility
between constructions and also within a given
construction type:
a green pepper ≈ a pepper which is green
a red herring 6= a herring which is red
the night is young 6= the young night
I handed in the paper = I handed the paper in
Kim ran over the dog ≈? Kim ran the dog over
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
36 INTRODUCTION
Challenges Posed by MWEs
• Syntactic idiomaticity
• Semantic idiomaticity
• Pragmatic idiomaticity
• Statistical idiomaticity
• Flexibility
• Productivity
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
37 INTRODUCTION
Productivity
• Varying level of productivity for different MWEs:
ad/post/*pre/*in/*apple/... hoc
call/ring/phone/*telephone up
Melbourne train driver, human language technology,
apple juice seat, ...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
38 INTRODUCTION
Motivation: Why Multiword Expressions?
• Pervasiveness in language
? MWEs estimated to be equivalent in number to
simplex words in mental lexicon
• Volatility (domain tuning, terminology, ...)
? axis of evil, make the pie higher, private equity, ...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
39 INTRODUCTION
• Challenge to NLP systems
? language understanding
? fluency
? robustness
• Nice interaction between linguistics, statistics and
computational linguistics
• MWEs in language learning environments
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
40 INTRODUCTION
• Lots of interesting crosslingual commonalities/divergences
? lexical equivalence: in the red vs. no vermelho
? structural equivalence: in the black vs. no azul
? semantic equivalence: in a corner vs. encurralado
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
41 INTRODUCTION
Computational Tasks/Issues
• Parsing/identification
• Extraction
• Syntactic classification
• Semantic classification
• Representation
• Crosslingual approaches to MWEs
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
42 INTRODUCTION
MWEs in NLP Applications
• IR (N-grams)
? phrase-based retrieval: mixed results (Salton and
Smith 1990; Lewis and Croft 1990)
? query expansion: mixed results (Mandala et al. 2000)
? compound nominals more effective than simplex
nominals as index terms (Wacholder and Song 2003)
• Tagging
? virtuous circle between MWE identification and
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
43 INTRODUCTION
tagging accuracy (Piao et al. 2003)
• Parsing
? MWEs account for 8% of parsing errors with precision
grammar (Baldwin et al. 2004)
? perfect knowledge of adverbial MWEs shown to
enhance parser accuracy
• Information extraction
? collocations used extensively in IE tasks (Lin 1998b)
• Machine translation
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
44 INTRODUCTION
? MWEs integral component of symbolic MT systems
(Gerber and Yang 1997; Bond and Shirai 1997)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
45 INTRODUCTION
Summary
• What is an MWE?
• What properties are associated with MWEs?
• Why are MWEs challenging for NLP?
• What NLP applications do MWEs feature in?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
46 INTRODUCTION
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
47 COMPUTATIONAL SYNTAX OF MWEs
COMPUTATIONAL SYNTAX OF
MWEs
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
48 COMPUTATIONAL SYNTAX OF MWEs
Case Study: English Resource Grammar
• HPSG-based linguistically-precise open-source grammar
• Compositional semantics based on MRS
• Reversible (parsing and generation)
• Medium coverage
• 8,218 types and 10,625 lexical entries (v06-jun-03)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
49 COMPUTATIONAL SYNTAX OF MWEs
English Resource Grammar (ERG) in Action
• Leave the report on the desk
<h1,e2:PRESENT*:NO_ASPECT*:MOOD:BOOL,
{h1:imp_m_rel(h3),
h4:pronoun_q_rel(x6:2PER:REAL_GENDER:ZERO_PRON:-*, h5, h7),
h8:pron_rel(x6),
h9:_leave_rel(e2, x6, x11:-:REAL_GENDER:3SG*, v10),
h12:_def_q_rel(x11, h14, h13),
h15:_report_rel(x11, v16:BOOL),
h15:_on_rel(e17:BOOL:NO_TENSE:ASPECT:MOOD, x11, x18:REAL_GENDER:3SG*:-),
h19:_def_q_rel(x18, h21, h20),
h22:_desk_rel(x18, v23:BOOL)},
{h3 QEQ h9,
h5 QEQ h8,
h14 QEQ h15,
h21 QEQ h22}>
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
50 COMPUTATIONAL SYNTAX OF MWEs
Basic Syntactic Approach
• Classify different MWE types according to their
syntactic flexibility and productivity, and determine the
appropriate analysis accordingly
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
51 COMPUTATIONAL SYNTAX OF MWEs
MWE Types
multiword expression
lexicalized phrase institutionalized phrase
fixed expression semi-fixed expression syntactically-flexible expression
non-decomposable idiomcompound nominal
proper name...
verb-particle constructionlight verb construction...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
52 COMPUTATIONAL SYNTAX OF MWEs
Fixed Expressions
multiword expression
lexicalized phrase institutionalized phrase
fixed expression semi-fixed expression syntactically-flexible expression
non-decomposable idiomcompound nominal
proper name...
verb-particle constructionlight verb construction...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
53 COMPUTATIONAL SYNTAX OF MWEs
Fixed Expressions
• by and large, in short , kingdom come, every which
way , ad hoc (cf. ad nauseum, ad libitum, ad
hominem,...), Palo Alto (cf. Los Altos, Alta Vista,...),
etc.
• Fixed string which undergoes neither morphosyntactic
variation (*in shorter) nor internal modification (*in
very short)
• Simple words-with-spaces representation is sufficient
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
54 COMPUTATIONAL SYNTAX OF MWEs
Fixed Expressions: Analysis
• Lexical entry for ad hoc:
ad_hoc_1 := intr_adj_le &
[ STEM < "ad", "hoc" >,
SEMANTICS [KEY ad_hoc_rel ]].
• Allows very ad hoc, but not *ad very hoc.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
55 COMPUTATIONAL SYNTAX OF MWEs
Semi-fixed Expressions
multiword expression
lexicalized phrase institutionalized phrase
fixed expression semi-fixed expression syntactically-flexible expression
non-decomposable idiomcompound nominal
proper name...
verb-particle constructionlight verb construction...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
56 COMPUTATIONAL SYNTAX OF MWEs
Semi-fixed Expressions
• kick the bucket, prostrate oneself, part of speech, San
Francisco 49ers
• Adhere to strict constraints on word order and
composition
• BUT undergo some lexical variation, e.g.:
? inflectional: kick/kicks/kicking/kicked the bucket
? reflexive pronominal: prostrate him/.../herself
? determiner selection: the/those 49ers
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
57 COMPUTATIONAL SYNTAX OF MWEs
Semi-fixed Expressions: Analysis
• Treat as word complex which is lexically variable atparticular positions:
part_of_speech_1 := n_intr_le &
[ STEM < "part", "of", "speech" >,
INFL-POS "1",
SEMANTICS [KEY part_of_speech_rel ]] & / part_n1.
kick_the_bucket := v_unacc_le &
[ STEM < "kick", "the", "bucket" >, INFL-POS "1",
SEMANTICS [KEY kick_the_bucket_i_rel ]] &
/ kick_v_np*_trans_le.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
58 COMPUTATIONAL SYNTAX OF MWEs
U.S. Sports Team Names
• the (Oakland) Raiders
• an/the [[(Oakland)Raiders ] player ]
• the [Raiders and 49ers ].
• the league-leading (Oakland) Raiders.
• an [[(Oakland) Raider ] spokesman ]
• *the Oakland 49ers
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
59 COMPUTATIONAL SYNTAX OF MWEs
U.S. Sports Team Names: Analysis
• Name:[
SPR / 〈 〉]
•
USTeamName:
SPR
⟨
Det[
definite]
⟩
NUM / plural
• oakland raiders 1 := USTeamName &[
LEX-SIGNS / 〈 oakland 1, raiders 1 〉SEMANTICS 〈 oakland raiders rel 〉
]
.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
60 COMPUTATIONAL SYNTAX OF MWEs
Compound Nouns
• Fully productive = any sequence of nouns can combine
to form a MWE (within pragmatic bounds)
• Underspecified semantic relation between the noun
modifier and head:
newspaper selection
school bus
orange juice seat
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
61 COMPUTATIONAL SYNTAX OF MWEs
Compound Nouns: Analysis
• Constructional analysis: N → N N
• Underspecified compound rel relation between nominal
elements, e.g. cardboard box :
cardboard rel(x)∧box rel(y)∧compound rel(x, y)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
62 COMPUTATIONAL SYNTAX OF MWEs
Syntactically-flexible Expressions
multiword expression
lexicalized phrase institutionalized phrase
fixed expression semi-fixed expression syntactically-flexible expression
non-decomposable idiomcompound nominal
proper name...
verb-particle constructionlight verb construction...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
63 COMPUTATIONAL SYNTAX OF MWEs
Syntactically-flexible Expressions
• write up, let the cat out of the bag , have a shower, ...
• Variable level of flexibility for different expressions
• Basic mechanism of lexical selection
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
64 COMPUTATIONAL SYNTAX OF MWEs
Verb-Particle Constructions
• Verb-Preposition Combinations:
? It was like falling off a log/*falling a log off .
? They fell quietly off the log .
? [Off how many logs] did the drunk fall?
• Verb-Particle Combinations:
? They wrote up the memo/wrote the memo up
? *Up how many memos did they write?
? *They wrote quietly up the memos.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
65 COMPUTATIONAL SYNTAX OF MWEs
Verb-Particle Constructions
• Compositional: write up, eat/gobble up
(activity → accomplishment)
• Noncompositional: look up, throw up
• write up the memo/write the memo up
look up the answer/look the answer up
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
66 COMPUTATIONAL SYNTAX OF MWEs
Verb-Particle Constructions: Analysis
• Verb selects for particle:
hand_out_v1 := v_particle_np_le &
[ STEM < "hand" >,
SEMANTICS [ KEY hand_out_rel,
--COMPKEY out_rel ] ].
• Assume “joined” word order is canonical (Dehe 2002)
and derive “split” word order by way of lexical rule
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
67 COMPUTATIONAL SYNTAX OF MWEs
Verb-Particle Constructions: Productivity
• For fully/semi-productive verb-particles, avoid
enumeration through use of lexical rules
• E.g., movement verb + directional particle:
run/walk/... up/down/around/in/...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
68 COMPUTATIONAL SYNTAX OF MWEs
Light Verbs
• Idiosyncrasy:
make a mistake, *do a mistake, *give a mistake
give a demo, do a demo, *make a demo
• Flexibility:
How many demos did Kim give?
...give a revealing demo
A demo was given.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
69 COMPUTATIONAL SYNTAX OF MWEs
Light Verbs
• make v1 := v lite l &[
STEM 〈 “make” 〉COMPLEMENTS 〈 NP[KEY make arg rel ] 〉
]
.
make arg rel
mistake rel argument rel ...
• make a mistake/error/boo-boo...
• make an argument/point/statement ...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
70 COMPUTATIONAL SYNTAX OF MWEs
Decomposable Idioms
• take advantage (of), pull strings, keep tabs on
• Flexibility:
They regretted the considerable advantage that had
been taken of the unfortunate situation.
Strings had been pulled to get Sandy the job.
The FBI kept closer tabs on Kim than they kept on
Sandy.
• Flexibility is highly variable.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
71 COMPUTATIONAL SYNTAX OF MWEs
Decomposable Idioms: Analysis
• cat out of bag :=
SEMANTICS
[
cat i rel
INST x
]
,
[
bag i rel
INST y
]
,
out rel
ARG1 x
ARG2 y
. . .
.
• cat i :=[
SEMANTICS 〈[
cat i rel]
〉]
& / cat n1.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
72 COMPUTATIONAL SYNTAX OF MWEs
• bag i :=[
SEMANTICS 〈[
bag i rel]
〉]
& / bag n1.
• Use root conditions to ensure that all elements are
present:[
cat i(x) ∧ bag i(y) ∧ out(x, y)]
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
73 COMPUTATIONAL SYNTAX OF MWEs
Institutionalized Phrases
multiword expression
lexicalized phrase institutionalized phrase
fixed expression semi-fixed expression syntactically-flexible expression
non-decomposable idiomcompound nominal
proper name...
verb-particle constructionlight verb construction...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
74 COMPUTATIONAL SYNTAX OF MWEs
Representing institutionalized phrases
• Store matrix of dependency pairs, with the (smoothed)
corpus-based frequency of each
• Statistically disprefer rather than symbolically rule out
certain word combinations
• Principal use in generation
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
75 COMPUTATIONAL SYNTAX OF MWEs
Simple! ... or then Again?
• To date, we have proposed 4 basic analyses and
categorised constructions according to the best fit with
those 4 analysis types
• Not all constructions are this compliant!
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
76 Test Case: Determinerless PPs
Test Case:
Determinerless PPs
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
77 Test Case: Determinerless PPs
Definition
• Determinerless PPs (PP−Ds) are MWEs comprising a
preposition (P) and a singular noun (NSing) without a
determiner:
Institutional Media Metaphor Temporal Means/Manner
at school on film on ice at breakfast by car
in church on TV at large on holiday by train
in gaol to video at hand on break by hammer
on campus off screen at leave by night by computer
at temple in radio at liberty by day via radio
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
78 Test Case: Determinerless PPs
Crosslinguistic Occurrence of PP−Ds
• Most languages with articles have PP−Ds
• Same semantic types attested in English, Albanian,
Tagalog, German, et al. (Himmelman 1998):
? Institution/Location: at school
? Metaphor/Abstract: at large
? Temporal: in winter
? Means/Manner: by car
• Focus of our research on English and Dutch
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
79 Test Case: Determinerless PPs
Corpus- and Lexicon-Occurrence of PP−Ds
• PP−Ds described statically in COMLEX and
WordNet, but account for only around 30% and
15%, respectively, of the token occurrences of PP−Ds
occurring ≥20 times in the BNC
• ≈0.3% of words in BNC are PP−Ds
• ≈0.2% of parse errors over a sample of the BNC caused
by syntactically-marked PP−Ds
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
80 Test Case: Determinerless PPs
The Syntax of PP−Ds
• Variability in syntactic markedness, productivity and
nominal modifiability for different PP−D constructions
• Non-productive, non-modifiable PP−Ds: ex cathedra,
ad hominem, ad nauseum
• Fully-productive, highly-modifiable PP−Ds: per
recruited student that finishes the project
• Most PP−Ds lie between these two extremes
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
81 Test Case: Determinerless PPs
Syntactic Markedness
• Syntactically-unmarked PP−Ds: NSing is uncountable
E.g. Institutions: in school, in gaol , but *in office
(cf. school is over vs. *office is over)
• Syntactically-marked PP−Ds: NSing is strictly countable
E.g. PPs headed by per : per person, but *per
information
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
82 Test Case: Determinerless PPs
Nominal Modifiability
• No modification: in *mental/*small hospital
• Idiosyncratic modification: at long/*lengthy/*short
last
• Relatively free modification: at great/considerable/tedious
length
• Modification seldom unrestricted, except in productive
constructions
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
83 Test Case: Determinerless PPs
Empirical Analysis of ModifiabilityDivergence
PP D(PP ||PP ) D(PP ||NP )
on horseback 0.00 0.04
before dawn 0.00 0.16
to hospital 0.02 0.32
up front 0.03 0.26
on record 0.10 0.76
in diameter 0.14 0.54
in school 0.18 0.26
on loan 0.18 0.71
by decree 1.62 2.07
on analysis 4.29 2.81
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
84 Test Case: Determinerless PPs
Modification Types
Obligatory Optional Impossible
at ∗(eye) on (summer)Noun
level vacation
at ∗(long) in (sharp) on (∗very)Adjective
range contrast top
at ∗(company) in (family)
expense courtEither
at ∗(considerable) in (open)
expense court
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
85 Test Case: Determinerless PPs
Coordination quirks
• Coordinate constructions: from mother to child , room
by room.
• Partial selectional mismatches: in brush and ink
• Full selectional mismatches: over mens en wereld
“about human being and world”, van stadion en hotel
“of stadium and hotel”
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
86 Test Case: Determinerless PPs
The semantics of PP−Ds
• All PP−Ds show a certain degree of (generally
systematic) semantic markedness on the noun:
? institutional: at school
? metaphoric: on ice
? generic uses: by car
• Some semantic systematicity to the prepositions
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
87 Test Case: Determinerless PPs
PP−Ds with institutional nouns
• Activity enrichment: in gaol “while being a prisoner”
and in school “while attending school”, cf. in a/the
gaol
• Familiarity enrichment: John is in town “John is in
(my/his/this) town”, cf. in a/the town
• Overlap between the two: at work “while at (my/his)
work/working”
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
88 Test Case: Determinerless PPs
Metaphorical PP−Ds
• Examples:
? English – at large, on ice, at last
? Dutch – op zak “in pocket/possession”, aan zet “(lit.)
on turn”
• Non-compositional, but some degree of morphological
systematicity in English: lastly/at last , edgy/on edge
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
89 Test Case: Determinerless PPs
PP−Ds with generic readings
• Examples: by car , by hand , via email
• Means/manner semantics of noun rare in subject/object
position
• Resist referential uses and familiarity enrichments, but
allow generic and activity readings:
I travelled to San Francisco by car. They’re/It’s a
great way to travel/#It rattled a lot.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
90 Test Case: Determinerless PPs
Analysis 1: fixed expression
• Word-with-spaces analysis for fully lexicalised PP−Ds:
at large, on track
• Prevents nominal modification, coordination
• Effective at capturing syntactically- and semantically-
marked PP−Ds
• Hard to capture optional preposition selection properties
(e.g. on top (of), in front (of)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
91 Test Case: Determinerless PPs
Analysis 2: simple combination
• Use head-complement rule to combine simplex P and N
lexical entries:
at church, in/after/during/... school
• Effective at capturing productive syntactically-unmarked
PP−Ds
• Licenses unconstrained nominal modification
• Captures compositional semantics
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
92 Test Case: Determinerless PPs
NP
PP
P
at
school
N
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
93 Test Case: Determinerless PPs
Analysis 3: N selection
• License the preposition to select for an unsaturated NP
(N):
by train/plane/bus/hydrofoil/pogo stick...
• Allows for nominal modification and full productivity
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
94 Test Case: Determinerless PPs
PP
P
by
car
N
N
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
95 Test Case: Determinerless PPs
Analysis 4: idiosyncratic modification
• Use idiomatic nominal lexical entries and unary rules to
constrain the nominal modifier type (noun, adjective or
neither):
at *(eye) level, on summer vacation, on (*very)
top.
• Specify preposition–noun combinatorics by way of root
conditions (a la decomposable idioms)
• Captures idiosyncratic modification effects
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
96 Test Case: Determinerless PPs
Determining the appropriate analysis
• Consider:
? nominal modifiability
? productivity
? semantic markedness
? NP saturation (syntactic markedness)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
97 Test Case: Determinerless PPs
An Analytical Challenge
• Some strictly-countable nouns occur prolifically in
PP−Ds without marked semantics, e.g.:
hand , and Dutch zee “sea” and huis “house”
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
98 Test Case: Determinerless PPs
Exercise: Pick the Right AnalysisAnalysis
MWEFixed Semi-fixed Syn-flex Inst Other
push up the daisiesbreak the icein timepecking orderfrom scratchbone of contentionshort shriftmake short work oftrue candourblow one’s toprun up
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
99 Test Case: Determinerless PPs
Summary
• What basic syntactic types of MWE are there?
• What issues do the various analyses address?
• How do we determine the correct analysis for a given
MWE?
• Are MWEs really that well behaved?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
100 Test Case: Determinerless PPs
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
101 EXTRACTING MWEs
EXTRACTING MWEs
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
102 EXTRACTING MWEs
Basic Task Description
• Identify the multiword expression (MWE) types in
tagged (or raw) text from observation of the token
distribution
• MWE (token) identification vs. (type) extraction
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
103 EXTRACTING MWEs
In American romance , almost nothing rates higher than what the movie men have called
“ meeting cute ” – that is , boy-meets-girl seems more adorable if it does n’t take place
in an atmosphere of correct and acute boredom . Just about the most enthralling real-life
example of meeting cute is the Charles MacArthur-Helen Hayes saga : reputedly all he did
was give her a handful of peanuts , but he said simultaneously , “ I wish they were emeralds
” . Aside from the comico-romantico content here , a good linguist-anthropologist could
readily pick up a few other facts , especially if he had a little more of the conversation to go
on . The way MacArthur said his line – if you had the recorded transcript of a professional
linguist – would probably have gone like this : A[fj] Primary stresses on emeralds and wish ;
; note pitch 3 ( pretty high ) on emeralds but with a slight degree of drawl , one degree of
oversoftness . Conclusions : The people involved ( and subsequent facts bear me out here
) knew clearly the relative values of peanuts and emeralds , both monetary and sentimental
. And the drawling , oversoft voice of flirtation , though fairly overt , was still well within
the prescribed gambit of their culture . In other words , like automation machines designed
to work in tandem , they shared the same programming , a mutual understanding not only
of English words , but of the four stresses , pitches , and junctures that can change their
meaning from black to white . At this point , unfortunately , romance becomes a regrettably
small part of the picture ; ; but consider , if you can bear it , what might have happened
if MacArthur , for some perverse , undaunted reason , had made the same remark to an
Eskimo girl in Eskimo . To her peanuts and emeralds would have been just so much blubber
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
104 EXTRACTING MWEs
In American romance , almost nothing rates higher than what the movie men have called
“ meeting cute ” – that is , boy-meets-girl seems more adorable if it does n’t take place
in an atmosphere of correct and acute boredom . Just about the most enthralling real-life
example of meeting cute is the Charles MacArthur-Helen Hayes saga : reputedly all he did
was give her a handful of peanuts , but he said simultaneously , “ I wish they were emeralds
” . Aside from the comico-romantico content here , a good linguist-anthropologist could
readily pick up a few other facts , especially if he had a little more of the conversation to go
on . The way MacArthur said his line – if you had the recorded transcript of a professional
linguist – would probably have gone like this : A[fj] Primary stresses on emeralds and wish ;
; note pitch 3 ( pretty high ) on emeralds but with a slight degree of drawl , one degree of
oversoftness . Conclusions : The people involved ( and subsequent facts bear me out here
) knew clearly the relative values of peanuts and emeralds , both monetary and sentimental
. And the drawling , oversoft voice of flirtation , though fairly overt , was still well within
the prescribed gambit of their culture . In other words , like automation machines designed
to work in tandem , they shared the same programming , a mutual understanding not only
of English words , but of the four stresses , pitches , and junctures that can change their
meaning from black to white . At this point , unfortunately , romance becomes a regrettably
small part of the picture ; ; but consider , if you can bear it , what might have happened
if MacArthur , for some perverse , undaunted reason , had made the same remark to an
Eskimo girl in Eskimo . To her peanuts and emeralds would have been just so much blubber
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
105 EXTRACTING MWEs
Complications in MWE Extraction
• Working out the extent of the collocation (phrase
boundary detection)
trip the light �
trip the light fantastic �
trip the light fantastic at �
• Fine line between collocations and simple default lexical
combinations
buy a car/purchase power
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
106 EXTRACTING MWEs
Statistical Tests Commonly Used
• Simple frequency: f(x, y)
• Pointwise/specific mutual information: log P (x,y)P (x)P (y)
• Dice’s coefficient: 2 f(x,y)f(x)f(y)
• (Student’s) t score
• (Pearson’s) chi-square (χ2)
• Z score
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
107 EXTRACTING MWEs
• Log likelihood
• Selectional association
...
Finding of Evert and Krenn (2001) that simple
frequency is as good as a wide range of collocation
extraction measures over German Adj-N and P-N-V
triple extraction tasks
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
108 EXTRACTING MWEs
Bigram Results from the WSJ
Rank Frequency Mutual information χ2 t test
1 of the Quadi Doum Posse Comitatus of the
2 in the Wrongful Discharge LORIMAR TELEPICTURES in the3 to NUMB Seh Jik Petits Riens to NUMB
4 for the Noo Yawk Wrongful Discharge on the5 to the WESTDEUTSCHE LANDESBANK Tupac Amaru the company6 of NUMB Naamloze Vennootschap Sary Shagan about NUMB
7 on the Caisses Regionales Outlaw Biker said it8 NUMB to Centenaire Blanzy GEMINI SOGETI for the
9 that the Guillen Landrau Centenaire Blanzy to be10 the company Ea Matsekha Smith-Corona Typewriters a share...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
109 EXTRACTING MWEs
Why Statistics?
• Pick up on word combinations which occur with
“significantly” high relative frequency when compared
to the frequencies of the individual words (i.e. f(x, y)
as compared to f(x) and f(y))
• Why so many different statistical tests?
? complications in evaluation (hard to say which is
the “best” test, conflicting results from different
researchers)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
110 EXTRACTING MWEs
? different corpora have different distributional
idiosyncracies
? different tests have different statistical idiosyncracies
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
111 Collocation Extraction: Xtract (Smadja 1993)
Collocation Extraction: Xtract
(Smadja 1993)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
112 Collocation Extraction: Xtract (Smadja 1993)
Outline
• Automatic method for extracting collocations from raw
text based on n-gram statistics
• Basic intuition: collocations are more rigid
syntactically and more frequent than other word
combinations
• Method: attempt to capture this intuition using the
basic statistics of word combinations
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
113 Collocation Extraction: Xtract (Smadja 1993)
Stage 1: Extract Significant Bigrams
• w and wi co-occur (wi is a collocate of w) if they are
found in a single sentence separated by fewer than 5
words
• a bigram (w, wi) is significant iff:
? w and wi co-occur more frequently than chance
? w and wi appear in a relatively rigid configuration
• Divide up the set of collocates according to POS
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
114 Collocation Extraction: Xtract (Smadja 1993)
Example Corpus
... multiword(−1) expressions ...
... multiword(−1) expressions ...
... dialect(−1) expressions ...
... dialect(−2) and expressions ...
... expressions of interest(2) ...
... multiword(−1) expressions ...
... collocation extraction ...
... expressions dialect(1) ...
... multiword(−1) expressions ...
... multiword(−1) expressions ...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
115 Collocation Extraction: Xtract (Smadja 1993)
Noun Co-occurrence Table
w = expressions, D = {−2,−1, 1, 2}multiword dialect collocation interest
collocatew1 w2 w3 w4
p−2i 0 1 0 0
p−1i 5 1 0 0
p1i 0 1 0 0
p2i 0 0 0 1
freq i 5 3 0 1
wi ∈ C yes yes no yes
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
116 Collocation Extraction: Xtract (Smadja 1993)
Statistics of Expectation
• f =∑
wi∈C freqi
|C| = 5+3+13 = 3 (frequency average)
• σ =
√
∑
wi∈C(freqi−f)2
|C| =
√
(5−3)2+(3−3)2+(1−3)2
3 ≈ 1.63
• ki = freqi−f
σ(strength)
• pi =∑
j∈D pji
|D| (pair count average)
• Ui =∑
j∈D(pji−pi)
2
|D| (pair count variance)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
117 Collocation Extraction: Xtract (Smadja 1993)
Collocation Filters
• Strength: ki > kα(= 1)
� select frequent collocates
• Spread: Ui > U0(= 1)
� select spiky distributions
• Peakiness: pji ≥ pi + (kβ ×
√Ui) kβ = 0.51
� identify interesting spikes
1Value of 10 suggested for kβ in Smadja (1993)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
118 Collocation Extraction: Xtract (Smadja 1993)
Back to our Example: Strength
• w1 (multiword)
? k1 = 5−31.63 = 1.22 > 1 �
• w2 (dialect)
? k2 = 3−31.63 < 1 �
• w4 (interest)
? k3 = 1−31.63 < 1 �
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
119 Collocation Extraction: Xtract (Smadja 1993)
Spread and Peakiness
• w1 (multiword)
? p1 = 0+5+0+04 = 1.25
? U1 = (0−1.25)2+(5−1.25)2+(0−1.25)2+(0−1.25)2
4 ≈ 20.31 > 1 �
pi + (kβ ×√
Ui) 1.25 + (0.5 ×√
20.31) ≈ 3.50
p−21 0 < 3.50 �
p−11 5 ≥ 3.50 �
p11 0 < 3.50 �
p21 0 < 3.50 �
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
120 Collocation Extraction: Xtract (Smadja 1993)
Stage 2: Bigrams to N-grams
• Independent filter to detect larger N-grams
• Method: for each fixed-distance collocate (w, wji ),
extract out contiguous word sequences where
max(p(word[i])) > T (= 0.75)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
121 Collocation Extraction: Xtract (Smadja 1993)
Example
w = resistance, w−3i = path
Concordances from the BNC:
... trod the path(−3) of least resistance , ...
... finding the path(−3) of least resistance will ...
... along the path(−3) of least resistance .
... the safest path(−3) of least resistance through ...
... took the path(−3) of least resistance and ...
� the path of least resistance is a rigid noun phrase
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
122 Collocation Extraction: Xtract (Smadja 1993)
Reflections
• (At the time) groundbreaking research on collocation
extraction
• Not effective at extracting out low-frequency words
• Difficulties in evaluating the results of collocation
extraction (applies to this day)
• Difficulties in extracting non-contiguous (predicative)
collocations such as verb particles
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
123 Collocation Extraction: Xtract (Smadja 1993)
Linguistics in Collocation Extraction
• Apply statistical measures to (head) bigrams in a given
dependency relation (e.g. subject-verb)
? filters out stop words, produces “collocations” of pre-
defined type for direct use in parsing, etc
• Look beyond contiguous bigrams, to bigrams occurring
within a “collocational window” of fixed size (e.g. within
3-4 words of each other)
• Utilise linguistic qualities of collocations:
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
124 Collocation Extraction: Xtract (Smadja 1993)
? limited internal modifiability (applicable as a post-
filter)
? limited substitutability (contrast with anti-
collocations, e.g. (strong/*powerful) coffee)
? non-compositional semantics
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
125 Collocation Extraction: Xtract (Smadja 1993)
SubstitutabilityConceptLexicalisation
...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
126 Collocation Extraction: Xtract (Smadja 1993)
Substitutability
• Most immediate means of testing substitutability via
synonyms (Pearce 2001b)
• Synonyms accessible from thesauri, but word sense
disambiguation is generally needed to isolate which
synset(s) over which to apply substitution test
• Possibilities of getting at synonyms via distributional
analysis (possibly based on dependency pairs)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
127 Collocation Extraction: Xtract (Smadja 1993)
Collocation Extraction and Evaluation
• Difficulties in evaluation collocation extraction
techniques due to lack of gold-standard datasets (what
is MWE?)
• Precision generally evaluated according to pre-compiled
LR or relative to corpus
• How to evaluate recall?
• How much is good enough?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
128 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Verb-particle Extraction (Baldwin
and Villavicencio 2002;
Baldwin (to appear))
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
129 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Verb-particle Constructions (VPCs)
• VPC = verb + obligatory particle(s)
? intransitive:
Kim calmed down v particle le
? transitive:
Kim handed in the paper
Kim handed the paper inv particle np le
Kim gets Sandy down v np prep particle only le
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
130 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Linguistic Properties of VPCs
• Transitive VPCs undergo the particle alternation (hand
in the paper vs. hand the paper in)
• With transitive VPCs, pronominal objects must be
expressed in the split configuration (hand it in vs.
*hand in it)
• Manner adverbs cannot occur between the verb and
particle (*hand it promptly in)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
131 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Extracting VPCs: Task Description
• Extract out full list of VPCs attested in a given corpus
(cf. generation of independent list of VPCs)
• Make no assumptions about corpus annotation (use only
information from pre-processors)
• Base extraction method on basic linguistic properties of
VPCs
• Develop technique to be robust over low-frequency
VPCs
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
132 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
The Joys of VPC Extraction
• Limited coverage of linguistic tests
• Variable word order
• Variable window length
• Structural/analytical ambiguity:
? hand [the paper] [in] [here] vs. hand [the paper] [in
here] vs. hand [the paper in here]
? hand [in] [the paper] vs. hand [in the paper]
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
133 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Corpus Analysis of VPCs
• Generate gold-standard VPC data by taking intersection
of VPCs in Alvey Tools data, COMLEX v3.0 and
ERG (total of 3,205 entries)
• Take random sample of 1,000 VPCs (1,577 LEs) and
manually check for occurrences of each lexical entry
(valence) in the Brown Corpus, WSJ and BNC
• Estimate the frequency of attested VPCs by voting
across a range of extraction methods (explained later)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
134 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
VPC Frequency Distribution
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
1 10 100 1000
% t
ypes
Corpus frequency
BNC
BrownWSJ
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
135 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Analysis of VPC Results
Corpus Attested LEs Ave. freq Median freq
Brown 21.4% 2.3 1
WSJ 21.2% 3.4 1
BNC 69.9% 89.6 7
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
136 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Why do We Need Extraction?
• Rather then extracting VPCs, why not just use
a pre-compiled broad-coverage, general-purpose VPC
dictionary?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
137 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Tasks
a. Shallow lexical acquisition: extraction of VPC types
without valence information (e.g. calm down, hand in)
b. Deep lexical acquisition: extraction of VPC
lexical entries (e.g. calm down = v particle le ∧v particle np le, blaze away = v particle le)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
138 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Evaluation
• Use standard measures of precision, recall and F-score
(β = 1)
• Calculate relative to the manually-determined corpus
attestations of the 1,000 VPCs (for each corpus), cast
in terms of the relative task
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
139 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Classifier design
a. Generate feature vectors based on various statistics of
VPC occurrence from training and test corpora, and
build classifier using TiMBL v4.2 (k-NN)
b. Evaluation according to 10-fold cross validation
• hold out test VPCs in training corpus data
• test corpus annotations only used in evaluation (not
in training data)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
140 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Training data
a. Corpus attestation data: manually-annotated corpus
attestation of 1,000 VPCs/1,577 LEs
b. Gold-standard dictionary data: 3,205 valence-
annotated VPC types (4,597 LEs)
• apply closed-world assumption in classifying VPC
training data
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
141 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-1: Simple POS-based Extraction
• Identify particles using dedicated POS tag (RP in Penn
and CLAWS2 tagsets)
• PROCEDURE:
a. tag the data using a tagger and lemmatise using morph
b. for each particle, search back to the left up to 6 words
to find governing verb
c. filter data according to set of 73 canonical particles
d. classify as transitive if split or immediately followed by
NP, otherwise intransitive
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
142 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-1: Feature Representation
• Features describing frequency of intransitive and
transitive VPC types:
INTRANS TRANS
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
143 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-1: Example
country NN fund NNS offer VBP an DT easy JJ
way NN to TO get VB a DT taste NN of IN
foreign JJ stock NNS without IN the DT hard JJ
research NN of IN seek VBG out RP individual JJ
company NNS . .
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
143 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-1: Example
country NN fund NNS offer VBP an DT easy JJ
way NN to TO get VB a DT taste NN of IN
foreign JJ stock NNS without IN the DT hard JJ
research NN of IN seek VBG out RP individual JJ
company NNS . .
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
144 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-1: Taggers
mxpost: Penn-based MaxEnt tagger
fnTBL: Penn-based TBL tagger
RASP: CLAWS2-based HMM tagger used in RASP
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
145 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-1: Results (Corpus-based)
Tagger Task Precision Recall Fβ=1
Shallow .973 .877 .922
mxpost Deepintrans .570 .447 .502
Deeptrans .886 .824 .854
Shallow .979 .825 .896
fnTBL Deepintrans .573 .447 .503
Deeptrans .894 .776 .831
Shallow .971 .735 .837
RASP Deepintrans .600 .525 .560
Deeptrans .834 .707 .765
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
146 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-1: Results (Dictionary-based)
Tagger Task Precision Recall Fβ=1
Shallow .973 .876 .922
mxpost Deepintrans .645 .658 .651
Deeptrans .871 .842 .857
Shallow .979 .822 .894
fnTBL Deepintrans .663 .627 .644
Deeptrans .856 .832 .844
Shallow .963 .537 .690
RASP Deepintrans .652 .451 .533
Deeptrans .829 .442 .577
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
147 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-1: Results
• Good results for the shallow task and transitive VPCs,
lesser so for intransitive VPCs
• mxpost and fnTBL roughly equivalent, RASP
significantly worse
• Corpus-based training data generally produces higher
precision, dictionary-based training data higher recall
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
148 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-2: Simple Chunk-based Extraction
• Identify particles using dedicated CoNLL-2000 chunk
tag (PRT)
• PROCEDURE:
a. chunk-parse tagged/lemmatised data using fnTBL
b. for each (canonical) particle, search back to the left
up to 6 words to find governing verb
c. only allow noun, preposition and adverb chunks
between verb and particle
d. valence determination similar to POS-based extraction
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
149 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-2: Feature Representation
• Features describing frequency of intransitive and
transitive VPC types:
INTRANS INTRANSL INTRANS% TRANS TRANSL TRANS% MI
where:
(IN)TRANSL = freq(linguistic test data)
(IN)TRANS = freq(other VPC instances)
(IN)TRANS% =freq((IN)TRANS)
freq(INTRANS)+freq(TRANS)MI = MI(V ;P )
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
150 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-2: Example
[O “] [PP instead of] [VP buy] [NP mask] [PP for]
[NP your kid] [O ,] [ADVP just] [VP cut] [PRT out]
[NP the columnist] [NP ’ picture] [O ...] [O .]
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
150 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-2: Example
[O “] [PP instead of] [VP buy] [NP mask] [PP for]
[NP your kid] [O ,] [ADVP just] [VP cut] [PRT out]
[NP the columnist] [NP ’ picture] [O ...] [O .]
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
151 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-2: Results
Training data Task Precision Recall Fβ=1
Shallow .991 .736 .845
Corpus Deepintrans .596 .489 .537
Deeptrans .936 .724 .817
Shallow .987 .735 .842
Dict Deepintrans .634 .614 .624
Deeptrans .881 .751 .811
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
152 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-2: Results
• Higher precision than Method-1, but recall goes down
considerably
cause: low chunk recall over particles
• Slightly disappointing results
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
153 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-3: Chunk Grammar-based
• Improve recall by looking also at canonical particles
occurring as non-particle (PP, ADV) chunks
• Use chunk grammar to determine the syntactic relation
between verbs and “particle candidates”
• Classify instances as:
? unambiguously intransitive/transitive VPC
? unambiguously intransitive/transitive non-VPC
? possible intransitive/transitive VPC
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
154 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-3: Identifying VPCs
• Use chunk grammar to:
? check that the chunks which occur between the verb
and particle are maximally an NP and particle pre-
modifier adverb chunk (back, right, ...)
? check for a clause boundary or NP immediately after
the particle/preposition/adverb chunk
? check the clause context of the verb chunk for possible
extraposition of an NP verbal complement
• Check congruity with linguistic properties of VPCs
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
155 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-3: Structural Ambiguity
[NP we] [VP may ask] [NP question] [SBAR as] [NP you]
[VP go] [ADVP along] [O ,] ... ✔
[NP it] [VP wo n’t do] [NP any good] [PP for]
[NP anybody] [SBAR unless] [NP employee] [VP know]
[PP about] [NP it] [O .] ✗
[VP nonperform] [NP loan] [VP will make] [PP up]
[NP only about 0.5 %] [PP of] [NP the combine bank]
[NP ’s total loan] [ADJ outstanding] ... ???
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
156 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-3: Attachment Disambiguation
• For cases of structural ambiguity, attempt to resolve
using log likelihood ratio (verb–particle (V P ), verb–NP1
head (V N1), NP1 head–particle (N1P ) and particle–NP2
head (PN2):
[VP hand] [NP1 the paper] [PP in] [NP2 here]
VPC realised iff:
V P × V N1 > N1P × PN2
V P × V N1 > V P × PN2
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
157 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-3: Feature Representation
• Features describing frequency of positive/negative
diagnostics for each (intrans/trans) VPC type:
INTRANS+ INTRANS− (INTRANSATT) TRANS+ TRANS− (TRANSATT)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
158 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-3: Results
Training data Task Precision Recall Fβ=1
Shallow .984 .891 .935
Corpus Deepintrans .670 .848 .748
Deeptrans .889 .821 .853
Shallow .982 .790 .876
Dict Deepintrans .753 .672 .710
Deeptrans .877 .762 .815
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
159 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-3: Results
• Appreciable gain in recall over Method-1 and Method-2
(greater robustness over low-frequency data)
• More credible results over deep processing tasks
• Corpus-based training data markedly better than
dictionary-based training data
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
160 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-4: Parser-based
• Use full parser to resolve attachment ambiguity
• Parser: RASP (tag sequence-based parser)
• Read VPCs off RASP output directly
• Valence determination directly from RASP output
(presence of dobj for head verb)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
161 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-4: Example (Full Parse)
(|He:1_PPHS1| |be+ed:2_VBDZ| |wound+ed:3_VVN| |,:4_,|
|but:5_CCB| |fight+ed:6_VVD| |on:7_RP|) 1 ; (-7.655)
(|ncsubj| |fight+ed:6_VVD| |He:1_PPHS1| _)
(|ncsubj| |wound+ed:3_VVN| |He:1_PPHS1| |obj|)
(|aux| _ |wound+ed:3_VVN| |be+ed:2_VBDZ|)
(|ncmod| _ |fight+ed:6_VVD| |on:7_RP|)
(|conj| _ |wound+ed:3_VVN| |fight+ed:6_VVD|)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
162 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-4: Example (Partial Parse)
... |to:15_TO| |bring:16_VV0| |out:17_RP|
|the:18_AT| |sheen:19_NN1|) 0 ; ()
(|ncsubj| |bring:16_VV0| |child+s:12_NN2| _)
(|dobj| |bring:16_VV0| |sheen:19_NN1| _)
(|ncmod| _ |bring:16_VV0| |out:17_RP|)
(|detmod| _ |sheen:19_NN1| |the:18_AT|)
(|xcomp| |to:15_TO| |hair:14_NN1| |bring:16_VV0|)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
163 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-4: Feature Representation
• Features describing frequency of each (intrans/trans)
VPC type, for full and partial parses:
INTRANSfull INTRANSpartial TRANSfull TRANSpartial
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
164 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-4: Results
Training data Task Precision Recall Fβ=1
Shallow .975 .715 .825
Corpus Deepintrans .632 .656 .644
Deeptrans .861 .705 .775
Shallow .975 .715 .825
Dict Deepintrans .643 .639 .641
Deeptrans .865 .705 .777
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
165 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method-4: Results
• Recall down as compared to Method-3
• Results superior to RASP tagger (parser pre-processor)
but below those of the other taggers
• Very little difference between corpus- and dictionary-
based training data
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
166 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Method Combination
• Combine methods to consolidate on relative strengths
(precision/recall)
• Combination by concatenating feature vectors for
individual methods
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
167 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Full Combination Results
Training data Task Precision Recall Fβ=1
Shallow .978 .957 .967
Corpus Deepintrans .638 .830 .721
Deeptrans .893 .890 .891
Shallow .979 .857 .914
Dict Deepintrans .660 .781 .715
Deeptrans .845 .796 .820
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
168 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Results of Method Combination
• System combination produces F-score superior to
individual methods
• Still disappointing results for intransitive VPCs
→ try selective system combination (chunker, chunk
grammar, RASP)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
169 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Selective Combination ResultsTraining data Task Precision Recall Fβ=1
Shallow .978 .942 .960
Corpus Deepintrans .651 .893 .753
Deeptrans .899 .874 .886
Shallow .979 .824 .895
Dict Deepintrans .651 .717 .683
Deeptrans .863 .754 .805
• Best F-score for intransitive VPCs (corpus-based
training data)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
170 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Full Combination Results (Brown)
Training data Task Precision Recall Fβ=1
Shallow .956 .888 .921
Corpus Deepintrans .783 .692 .735
Deeptrans .840 .766 .801
Shallow .973 .555 .706
Dict Deepintrans .675 .314 .429
Deeptrans .765 .571 .654
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
171 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Full Combination Results (WSJ)
Training data Task Precision Recall Fβ=1
Shallow .938 .916 .927
Corpus Deepintrans .664 .669 .667
Deeptrans .875 .812 .843
Shallow .983 .559 .713
Dict Deepintrans .561 .274 .368
Deeptrans .830 .612 .704
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
172 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Reflections
• A range of methods proposed for shallow/deep lexical
acquisition of VPCs from unannotated corpora
• Blurring of the token/type distinction used to boost
performance
• Method robust over extremely low-frequency data (vital
for many MWE types)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
173 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Predicting VPC Productivity
• Verb semantics are often a good predictor of verb–
particle combinatorics (for compositional VPCs)
• Try using Levin classes to predict productive verb–
particle combinations (e.g. aspectual up)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
174 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Levin-based Productivity (vs. Dictionaries)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
175 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Levin-based Productivity (vs. Dict + BNC)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
176 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
Summary
• What is the basis of collocation extraction methods?
• How do collocation and MWE extraction methods
differ?
• What properties of MWEs make collocation extraction
techniques unsuitable?
• In what way can MWE extraction circumvent the issue
of lexical coverage?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
177 Verb-particle Extraction (Baldwin and Villavicencio 2002; Baldwin (to appear))
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
178 MWE INTERPRETATION: COMPOUND NOUNS
MWE INTERPRETATION:
COMPOUND NOUNS
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
179 MWE INTERPRETATION: COMPOUND NOUNS
Compound Nominals and Nominalisations
• Compound nominal: N made up of two or more nouns,
e.g.:
telephone box/booth, river bed, radar footprint,
chest X-ray
• Nominalisation: subclass of compound nominals in
which the head noun is deverbal, e.g.:
machine performance, museum construction,
family worker, student education, satellite observation
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
180 MWE INTERPRETATION: COMPOUND NOUNS
Compound Nominals and NLP
• Compound nominals generally processed in three steps:
a. Identification of compound nominals in some corpus
A film interpretation of the book which satirises
black assimilation into white society.
b. Syntactic analysis of the structure
engine oil filter � [[engine oil] filter]
c. Interpretation of the semantics
film interpretation � obj
• We will focus on interpretation (Step 3)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
181 MWE INTERPRETATION: COMPOUND NOUNS
Identification
• Compound nominals easily detectable from the output
of a tagger:
A DT film NN interpretation NN of IN the DT
book NN which WDT satirises VBZ black NN
assimilation NN into IN white NN society NN
. .
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
181 MWE INTERPRETATION: COMPOUND NOUNS
Identification
• Compound nominals easily detectable from the output
of a tagger:
A DT film NN interpretation NN of IN the DT
book NN which WDT satirises VBZ black NN
assimilation NN into IN white NN society NN
. .
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
182 MWE INTERPRETATION: COMPOUND NOUNS
Identification
• Largely a question of POS tagger post-correction
(Lapata and Lascarides 2003)
• Subtle questions about how to detect compound
nominals which are part of larger lexical items (e.g.
social services committee)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
183 MWE INTERPRETATION: COMPOUND NOUNS
NN Corpus Occurrence
• Estimate of English and Japanese NN corpus occurrence:
BNC Reuters Mainichi
Token coverage 2.6% 3.9% 2.9%
Total no. types 265K 166K 889K
Ave. token freq. 4.2 12.7 11.1
Singletons 60.3% 44.9% 45.9%
• Highly productive, high frequency of occurrence
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
184 MWE INTERPRETATION: COMPOUND NOUNS
Syntactic Analysis
• Adjacency (Resnik 1993) vs. dependency (Lauer
1995a) in syntactic analysis, e.g. woman aid worker :
woman aid > aid worker [Adj]
aid worker > woman worker [Dep]
→ [[woman aid] worker]
woman aid < aid worker [Adj]
aid worker < woman worker [Dep]
→ [woman [aid worker]]
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
185 MWE INTERPRETATION: COMPOUND NOUNS
Interpretation
• Compound nominals are largely unrestricted
semantically
diesel truck/oil/tanker, phone book, cloud bus,
apple juice seat
• Nominalisations tend to occur with subject or object
interpretation:
machine performance, museum construction,
student education BUT ALSO soccer competition
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
186 MWE INTERPRETATION: COMPOUND NOUNS
Why Compound Nominal Interpretation?
• Component of language understanding
• (Partial) interpretation required for MT into certain
languages:
cf. Italian: coltello da pane “bread knife”, porta a
vetri “glass door”, succo di limone “lemon juice”
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
187 MWE INTERPRETATION: COMPOUND NOUNS
• Interface between semantics and stress assignment:
pint jar/mile run/six-figure salary/...
pantry shelf/garage door/bedroom furniture/...
wood box/water bucket/gin bottle/...
daisy chain/cable network/sugar cube/...
BUT rubber boots/steel plate/gold medal/...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
188 MWE INTERPRETATION: COMPOUND NOUNS
Interpreting Compound Nominals
• Possible to interpret compound by way of:
? system of “semantic relations”
activity, change, person-afflicted, ...
steel can = made-of(can,steel)
? paraphrasing with prepositional “hidden variable”
P (n2, p, n1) ≈ P (n2, p, ∗)P (∗, p, n1)
baby chair = chair for babies
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
189 MWE INTERPRETATION: COMPOUND NOUNS
Semantic Theories
• Every linguist has her own theory, but with
commonalities
• Import of syntax, semantics, discourse and knowledge
representation in different theories
• Claims that finite enumeration of semantic relations are
psychologically untenable (Downing 1977)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
190 MWE INTERPRETATION: COMPOUND NOUNS
Example Theory 1: Levi (1978)
• 4 roles for nominalisations:
? act, product, agent, patient
truck driver = agent
student discontinuation = act
• 9 recoverably deletable predicates for compound nominals:
? in, for, from, about (prepositional)
? cause, make, have, use, be (relative clauses)
power station = make
steel box = use
baby crocodile = be
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
191 MWE INTERPRETATION: COMPOUND NOUNS
Example Theory 2: Lauer (1995a)
• Interpret compound nominals according to 7 prepositions:
? of : state law = law of state
? for: baby chair = chair for baby
? in: morning prayer = prayer in morning
? at: airport food = food at airport
? on: Sunday television = television on Sunday
? from: reactor waste = waste from reactor
? with: gun man = man with gun
? about: war story = story about war
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
192 MWE INTERPRETATION: COMPOUND NOUNS
Example Theory 3: Copestake (2003)
• Cateogrise compounds as first category that “fits”:
a. listed compounds: home secretary
b. hypernmic compounds: tuna fish, oak tree
c. deverbal compounds: satellite observation
d. relational compounds: jazz fan
e. made-of compounds: steel sword, polystyrene box
f. prepositional compounds: airshow accident
g. non-deverbal verb compounds: oil town
h. non-paraphrasable compounds: listeria society
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
193 MWE INTERPRETATION: COMPOUND NOUNS
Exercise: Analyse the NNInterpretation
NNLevi Lauer Copestake
machine translationcold viruscardboard boxstate premiertax moduledisk cylinderrelaxation classdarts competitiontennis coachcity protesttelephone number
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
194 MWE INTERPRETATION: COMPOUND NOUNS
Open Questions
• Is there a definitive categorical system of compound
nominal interpretation types? (splitters and lumpers)
• Can any one system work for all domains and compound
nominal types?
• What systems of interpretation work in different
domains?
• To what degree is interpretation required?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
195 The Disambiguation of Nominalisations (Lapata 2002)
The Disambiguation of
Nominalisations (Lapata 2002)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
196 The Disambiguation of Nominalisations (Lapata 2002)
Basic Outline
• Task: binary classification of nominalisations as having
a subj or obj interpretation (ignore nominalisations
such as soccer competition — i.e. constrain the space
in such a way that interpretation is a well-defined task)
• Assumption: P (rel|n1, n2) ≈ P (rel|vn2, n1)
• Problem: getting accurate estimates of P (rel|vn2, n1)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
197 The Disambiguation of Nominalisations (Lapata 2002)
Basic Model
RA(rel, n1, n2) = log2
P (obj|n1, n2)
P (subj|n1, n2)
P (rel|n1, n2) ≈ f(vn2, rel, n1)∑
i f(vn2, reli, n1)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
198 The Disambiguation of Nominalisations (Lapata 2002)
Resources
• Derive frequency estimates from the BNC
• Estimate f(vn2, rel, n1) from output of dependency
parser (Cass)
• Determine base verb form of nominalisation based on
nomlex and celex
• Hand-annotate/filter 796 nominalisations extracted
from BNC
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
199 The Disambiguation of Nominalisations (Lapata 2002)
Observation
• Of 796 items in gold-standard nominalisation set, 47%
not attested in BNC in either a verb-object or verb-
subject relation
• How to get accurate estimates of f(vn2, rel, n1)?
• Answer: smoothing based on the frequencies of
observed verb-argument pairs
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
200 The Disambiguation of Nominalisations (Lapata 2002)
Smoothing
a. Discounting: redistribute probability from observed
events to unobserved events
b. Class-based smoothing: word-to-class distributional
similarity
c. Distance-weighted averaging: word-to-word
distributional similarity
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
201 The Disambiguation of Nominalisations (Lapata 2002)
Discounting
• Katz’s backing-off:
P (rel|n1, n2) =
αf(vn2,rel,n1)
� i f(vn2,reli,n1)if f(vn2, rel, n1) > 0
βf(rel,n1)
f(n1)if f(rel, n1) > 0
(1 − α − β) f(rel)
� i f(reli)otherwise
• Estimate α and β by Good-Turing estimation
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
202 The Disambiguation of Nominalisations (Lapata 2002)
Class-based Smoothing
• Map observed verb-argument tuples onto the
WordNet/Roget classes of the noun, distributing equally
across all synsets the noun is categorised as belonging
to
• Calculate f(vn2, rel, n1) by averaging across the classes
that n1 occurs in
• Closed world assumption for nouns
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
203 The Disambiguation of Nominalisations (Lapata 2002)
Distance-weighted Averaging
• Use confusion probability or Jensen-Shannon
divergence to estimate the distributional similarity
between vn2 and each verb w′1, and estimate
f(vn2, rel, n1) according to:
fs(vn2, rel, n1) =∑
w′1
sim(vn2, w′1)f(w′
1, rel, n1)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
204 The Disambiguation of Nominalisations (Lapata 2002)
Confusion Probability
PC(w1|w′1) =
∑
rel,w2
P (w1|rel, w2)P (rel, w2|w′1)
=∑
rel,w2
f(w1, rel, w2)
f(rel, w2)
f(w′1, rel, w2)
f(w′1)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
205 The Disambiguation of Nominalisations (Lapata 2002)
Jensen-Shannon Divergence
J(w1, w′1) =
1
2
[
D(
m(w1)||n(w1, w′1)
)
+ D(
m(w2)||n(w1, w′1)
)
]
WJ(w1, w′1) = 10−βJ(w1,w′
1)
where
m(w) = P (rel, w2|w)
n(w1, w′1) =
1
2
(
m(w1) + m(w′1)
)
D(
m(w1)||n(w1, w′1)
)
=
∑
rel,w2
P (rel, w2|w1) logP (rel, w2|w1)
12
(
P (rel, w2|w1) + P (rel, w2|w′1)
)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
206 The Disambiguation of Nominalisations (Lapata 2002)
Evaluation
• Annotator agreement = 89.7%
• Take 2,000 nearest neighbour verbs w′1 distance-
weighted averaging methods, β = 5
• Baseline accuracy of 61.5% (obj interpretation)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
207 The Disambiguation of Nominalisations (Lapata 2002)
Results
• Confusion probability and WordNet-based smoothing
tend to do the best overall
• Good results for system classification, combined with
context modelling in the form of the right word context
of the compound nominal (85% test accuracy)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
208 The Disambiguation of Nominalisations (Lapata 2002)
Reflections
• Interesting task-oriented smoothing experiment
• What to do with non-subj/obj nominalisations?
• What to do with prepositional verbs, verb particles?
• Influence of pragmatics on interpretation
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
209 Classifying the Semantic Relations in Noun Compounds (Rosario and Hearst 2001)
Classifying the Semantic Relations
in Noun Compounds (Rosario and
Hearst 2001)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
210 Classifying the Semantic Relations in Noun Compounds (Rosario and Hearst 2001)
Basic Outline
• Task: interpretation of (2-word) compound nominals
within the biomedical domain
• Method: use lexical or conceptual knowledge about
the component nouns to interpret the whole (context-
independent)
• Resource: MeSH (biomedical thesaurus)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
211 Classifying the Semantic Relations in Noun Compounds (Rosario and Hearst 2001)
Semantic Roles
• Compound nominals interpreted via 18 (out of 38)
relations:
? more specific than case roles, and less specific than IE
template fillers
? customised to the biomedical domain (e.g. polio
survivors � person-afflicted)
? thresholded for frequency
? overlapping (multiclass classification possible: cell
growth � activity + change)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
212 Classifying the Semantic Relations in Noun Compounds (Rosario and Hearst 2001)
Method
• Class-based model: describe NN according to the
concatenation of the MeSH representations of N1 and
N2 (up to level N)
• Lexical model: describe NN by its component words
(closed-word assumption)
• Learner: neural network (feed-forward network with
one hidden layer)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
213 Classifying the Semantic Relations in Noun Compounds (Rosario and Hearst 2001)
Results
• Over closed data, the lexical and class-based models
perform equivalently (≈ 60%)
• Over open data, the class-based model performs better
(unsurprisingly)
• Suggestion that N2 has a stronger impact on the
interpretation than N1
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
214 Classifying the Semantic Relations in Noun Compounds (Rosario and Hearst 2001)
Reflections
• Question of interpretation system sidestepped to some
degree by picking a technical domain
• Multiclassification awkward effect, which raises
questions about the appropriateness of the interpretation
system
• Possibility for a hybrid approach combining the class-
based and lexical models?
• No systematic treatment of lexicalised nominals
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
215 Integrating Symbolic and Statistical Representations (Copestake and Lascarides 1997)
Integrating Symbolic and Statistical
Representations (Copestake and
Lascarides 1997)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
216 Integrating Symbolic and Statistical Representations (Copestake and Lascarides 1997)
Basic Outline
• Basic method:
a. use the grammar/lexicon to delimit the range of
potential interpretations of a given NN
b. use “productivity” probabilities to rank the individual
interpretations
c. use pragmatics to filter out interpretations which
produce discourse incoherence within a given context
• Possible to derive non-standard interpretations for a
compound nominal (e.g. garbage man)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
217 Integrating Symbolic and Statistical Representations (Copestake and Lascarides 1997)
Semantic Hierarchy
n n rule������������������
PPPPPPPPPPPPPPPPPP
made-of purpose-patient@
@@
@@
@@
��
��
��
�
deverbal�
��
��
��
@@
@
cardboard box
deverbal-ppnon-derived-pp
linen chest ice-cream container
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
218 Integrating Symbolic and Statistical Representations (Copestake and Lascarides 1997)
Estimating Productivity
• Estimate productivity based on the number of attested
forms of a given schemata:
Prod(cmp schema) =M + 1
N
where N is the number of pairs of senses which match
cmp schema and M is the number of attested forms
• Cf. substitution tests for collocations/compositionality
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
219 Integrating Symbolic and Statistical Representations (Copestake and Lascarides 1997)
Applying the Productivity Estimates
• Interpretations for cotton bag based on analysis of
fabric/container NNs in the BNC (based on WordNet):
made-of P = 0.84
purpose-patient P = 0.14
general-nn P = 0.02
• Prediction that the default interpretation for cotton bag
is made-of
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
220 Integrating Symbolic and Statistical Representations (Copestake and Lascarides 1997)
Interface with Pragmatics
• Model pragmatics with SDRT and world knowledge with
DICE
• Use SDRT and DICE to filter out interpretations that
produce discourse incoherence:
a. Mary sorted her clothes into various bags
made from plastic
b. She put her skirt into the cotton bag
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
221 Integrating Symbolic and Statistical Representations (Copestake and Lascarides 1997)
Reflections
• Rare instance of method which provides direct handling
of the lexicon-pragmatics interface
• Implausible interpretations supported explicitly, but
dispreferred
• Difficulties in collecting productivity statistics
• Question of real-world applicability of SDRT/pragmatic
reasoning
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
222 Integrating Symbolic and Statistical Representations (Copestake and Lascarides 1997)
Summary
• What basic types of compound noun are there, and how
do they differ?
• What types of theory are there for interpreting
compound nouns?
• What are their strengths and weaknesses?
• What computational techniques can be employed to
interpret compound nouns?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
223 Integrating Symbolic and Statistical Representations (Copestake and Lascarides 1997)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
224 SEMANTIC COMPOSITIONALITY
SEMANTIC COMPOSITIONALITY
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
225 SEMANTIC COMPOSITIONALITY
Semantic Decomposability and
Compositionality
• Decomposability = degree to which the semantics of an MWE
can be ascribed to those of its parts
kick the bucket → die′
spill the beans → reveal′(secret′)
• Compositionality = degree to which the semantics of the parts of
an MWE contribute towards those of the whole
• Domain considerations: monosodium glutamate in chemistry vs.
health domains
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
226 SEMANTIC COMPOSITIONALITY
Syntactic Compositionality
• Degree to which the syntactic properties of the parts of
an MWE combine to make up the syntax of the whole
? Fixed expressions: by and large, San Francisco
? Verb particles: eat up vs. chicken out
• Syntactic compositionality binary effect; non-
compositional MWEs lexicalised
• Semantic decomposability continuum of regularity with
more subtle effects and syntactic corollaries
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
227 SEMANTIC COMPOSITIONALITY
Decomposability and Syntactic Flexibility
• Consider:
*the bucket was kicked by Kim
Strings were pulled to get Sandy the job.
The FBI kept closer tabs on Kim than they kept on Sandy.
... the considerable advantage that was taken of the situation
• The syntactic flexibility of an idiom can generally be
explained in terms of its decomposability
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
228 SEMANTIC COMPOSITIONALITY
Ideal Research Objective
• Automatically decompose a given MWE/align
component words with semantic primitives
• Classification of MWEs into 3 classes:
a. non-decomposable MWEs (e.g. kick the bucket,
shoot the breeze, hot dog)
b. idiosyncratically decomposable MWEs (e.g. spill
the beans, let the cat out of the bag , radar footprint)
c. simple decomposable MWEs (e.g. kindle
excitement , traffic light)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
229 SEMANTIC COMPOSITIONALITY
Realistic Short-term Objective
• Demarcate simple decomposable MWEs from
idiosyncratically decomposable and non-decomposable
MWEs (roughly equivalent to endocentric vs.
exocentric distinction)
• Binary distinction vs. mapping onto continuum of
relative decomposability
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
230 SEMANTIC COMPOSITIONALITY
Approaches to Evaluation
• Dictionary based: binary evaluation, based on
prediction that non-compositional MWEs will be
lexically listed
• Similarity based: relative similarity of the parts to the
whole (e.g. relative to WordNet)
sim(pig metal ,metal) � sim(pig metal ,pig)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
231 SEMANTIC COMPOSITIONALITY
• Entailment based: binary evaluation, based on whether
the whole “entails” the parts or not
Susan finished up her paper |= Susan finished her
paper
• Ranking based: describe MWE compositionality by
way of continuous/discrete scale of compositionality
comp(put up) ≥ comp(eat up) ≥ comp(gun down)
...
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
232 SEMANTIC COMPOSITIONALITY
Exercise: Rate the Compositionality
CompositionalityVPC
Dic Sim Ent(V) Ent(P) Rankget downtrans
piss offtrans
pay offtrans
lift outtrans
roll backtrans
dig uptrans
lie downintrans
wear onintrans
chicken outintrans
hand outtrans
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
233 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Automatic Identification of
Non-Compositional Phrases (Lin
1999)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
234 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Basic Method
• Use substitution as a test of compositionality:
red tape → yellow tape, red cassette
economic impact → political impact , economic
effect
• Evaluate based on a dictionary of idioms
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
235 Automatic Identification of Non-Compositional Phrases (Lin 1999)
System Resources
• POS-conditioned thesaurus (nouns, verbs, adjectives/adverbs)
? derived from dependency data (Minipar):
• Collocation data
? dependency tuples (H,R,M) with high log-likelihood
ratio (H = head, R = relation, M = modifier)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
236 Automatic Identification of Non-Compositional Phrases (Lin 1999)
(Point-wise) Mutual Information
• Measure of the level of association between two events
A and B:
MI(A, B) = log2
P (A, B)
P (A)P (B)
• Commonly used in collocation extraction
• Not appropriate for low-frequency events
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
237 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Mutual Information and Compositionality
• Scaling up to 3 events A, B and C, where B and C are
conditionally independent given A:
MI(A, B,C) = log2
P (A, B, C)
P (B|A)P (C|A)P (A)
MI(H, R, M) = log2
|H R M||* * *|
|H R *||* R *|
|* R M||* R *|
|* R *||* * *|
= log2
|H R M||* R *|
|H R *||* R M|
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
238 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Definition of Compositionality
• A phrase α is non-compositional iff there is no β s.t.:
(a) β can be produced by substitution of the components
of α for any of 10 most-similar words, and
(b) there is an overlap between the 95% confidence
interval of the MI values of α and β
• 10 most-similar words tested for each of H and M (R
fixed)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
239 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Word Similarity: Lin (1998)
sim(w1, w2) =∑
(r,w)∈T (w1)∩T (w2) MI(w1, r, w) + MI(w2, r, w)∑
(r,w)∈T (w1)MI(w1, r, w) +
∑
(r,w)∈T (w2) MI(w2, r, w)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
240 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Example 1: spill (one’s) guts
• (spill ,V:comp1:N,gut):
? spill : leak, pour, spew, ..., spray
? gut : intestine, instinct, foresight, ..., charisma
• Check for each of (leak ,V:comp1:N,gut), (spill ,V:comp1:N,intestine),
... in the collocation database
• None found, so spill (one’s) guts is non-compositional
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
241 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Example 2: red tape
• (tape,N:adj:N,red):
? tape: videotape, cassette, videocassette, ..., audio
? red : yellow, purple, pink, ..., shade
• Find (tape,N:adj:N,yellow), (tape,N:adj:N,orange),
(tape,N:adj:N,black) in the collocation database but
with very different MI values
• red tape is non-compositional
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
242 Automatic Identification of Non-Compositional Phrases (Lin 1999)
MI Confidence Interval: the Z-test
• Possible to calculate the “true” MI of (H,R,M) according
to the Z-test:
p ± zN
√
p(1 − p)
n= k
n± zN
√
kn(1−k
n)
n≈ k±zN
√k
n
where p is the MLE of p, n is |* * *|, k is |H R M|,
and zN is a constant determined by the confidence level
N , e.g. z0.95 = 1.96
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
243 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Applying the Z-test
• Determine the “fit” between two MI values by
calculating the Z-score interval for the putative non-
compositional MWE and determining whether the MI
of the second falls into that interval
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
244 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Evaluation
• Evaluate the method relative to an idiom dictionary
• OK precision, and significant numbers of the extracted
MWEs not contained in the dictionary appear to be
non-compositional based on manual inspection
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
245 Automatic Identification of Non-Compositional Phrases (Lin 1999)
Reflections
• Is substitution really a good test for non-
compositionality?
? institutionalised phrases: frying pan, salt and pepper ,
many thanks
? productive MWEs: call/phone/ring up
• Look to alternative methods
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
246 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
A Statistical Approach to the
Semantics of Verb-particles
(Bannard et al. 2003)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
247 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Basic Method
• Define similarity in terms of distributional similarity, i.e.
assume that if an MWE is compositional, it will occur
in the same lexical context as its parts
• Divide up compositionality to look at verb and particle
similarity independently
• Evaluate against human judgements
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
248 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Verb-particle constructions (VPCs)
• VPC = A verb plus one or more obligatory
(prepositional) particles
Peter put the picture up
Susan finished up her paper
Philip gunned down the intruder
Barbara and Simon made out
• Assumption: VPCs are not always fully compositional
or fully non-compositional, but rather populate a
continuum between the two extremes
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
249 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Compositionality by Entailment
• Peter put the picture up
|= Peter put the picture somewhere
|= the picture was up
• Susan finished up her paper
|= Susan finished her paper
• Philip gunned down the intruder
|= the intruder was down
• Barbara and Simon made out
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
250 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Obtaining Human Judgements
a. Extract VPCs from British National Corpus (Baldwin
and Villavicencio 2002)
b. Randomly select 5 sentence tokens for each of 40
randomly selected VPC types
c. Present native English speakers with tokens and asked
whether verb and/or particle is implied by the VPC
d. Response: Yes, No or Don’t Know
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
251 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Example Sentence Tokens: round up
A dog started to round up sheep.
In three years they had rounded up fifty captive orangs.
Owned by Jo Rutherford, Trigo rounded up the milking herd and
brought it back to the milking parlour in Devon.
On 9 August, 349 Arrests were made as the miltary swooped to
round up serving and former IRA activists.
Ten days later, when the agents moved in to round up their
targets, El-Jorr checked out and returned to Cyprus, charging the
hotel bill to his American Express card as instructed.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
252 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Human Judgements
• Does round up imply round?
• Does round up imply up?
• Obtain gold-standard analysis by taking majority
judgement (ignore Don’t Know responses)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
253 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Sample Judgements
VPC Component word Yes No Don’t Know
dig 21 5 0dig up
up 18 7 1
stay 20 5 1stay up
up 21 5 0
brighten 9 16 1brighten up
up 16 10 0
add 12 14 0add up
up 19 6 1
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
254 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Binary Classification Tasks (×4)
a. Is the item completely compositional?
b. Does the item include at least one item that is
compositional?
c. Does the verb contribute its simplex meaning?
d. Does the particle contribute its simplex meaning?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
255 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Classification Methods
• Four different classifiers implemented
• Method 1 based on Lin (1999), Methods 2-4 address
theoretical concerns with this model
• All methods based on co-occurrence vector
representation of VPC and component words
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
256 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Method 1
• Reimplemented Lin (1999) over VPCs.
• Tested over all four tasks - assuming that the
substitutability of each item will give us some insight
into its semantic contribution
• Reconstruct Lin’s thesaurus to include all of verbs,
nouns, adjectives/adverbs and prepositions.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
257 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Method 2
• Similar to Lin (1999) except for use of knowledge-free
approach to obtaining thesaurus
• Very similar to Schutze (1998) “context space” method
• Similarities from pairwise comparison of all verbs,
particles and VPCs
• Obtain thesaurus by taking the N most similar words of
a given POS
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
258 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Method 3
• Use same method of substitution
• A component is said to be contributing simplex meaning
if expression formed by substitution occurs among the
nearest 100 verb-particle constructions
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
259 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Method 4
• Hypothesis is that if a verb or particle is contributing
simplex meaning to a VPC then it will be semantically
similar to the VPC according to cosine measure
? a verb is judged to be contributing simplex meaning if
it occurs within the 20 most similar items to the VPC.
? a particle is judged to be contributing simplex meaning
if it is in top 10 most similar items to the VPC.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
260 A Statistical Approach to the Semantics of Verb-particles (Bannard et al. 2003)
Results
• Mixed at best!
• Methods 3 and 4 tend to perform better than methods
1 and 2
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
261 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
Detecting a Continuum of
Compositionality in Phrasal Verbs
(McCarthy et al. 2003)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
262 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
Basic Method
• Compare different methods for modelling compositionality
based on distributional similarity and statistical tests
traditionally used in collocation extraction
• Map 111 VPCs onto a ranked list, based on human
judgements over an 11-point compositionality scale
• Evaluate according to the rank order correlation with
the gold-standard ranked list
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
263 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
System Resources
• Build thesaurus a la Lin (1998), based on dependency
tuple output of RASP
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
264 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
Similarity-based Methods
• overlap: relative overlap between the top N neighbours
of the VPC and its simplex verb
• sameparticle: the number of VPCs which select for
the same particle as the given VPC amongst the top N
neighbours of that VPC
• sameparticle − simplex: the value for sameparticle
minus the number of top N neighbours of the simplex
verb which select for that same particle
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
265 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
• simplexasneighbour: does the simplex verb occur in
the top 50 neighbours of the VPC?
• rankofsimplex: what is the rank of the simplex verb in
the neighbours of the VPC?
• overlapS: the overlap of neighbours in the top N
neighbours of the VPC and simplex verb, where VPC
neighbours are converted to simplex verbs in the VPC
case
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
266 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
Statistical Methods
• χ2
• Log-likelihood ratio
• (Point-wise) mutual information
• Simple frequency of the VPC
• Simple frequency of the simplex verb
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
267 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
Resource-based Method
• Binary test for the occurrence of the VPC in:
? WordNet
? Alvey Tools (ANLT) VPC data
? Alvey Tools (ANLT) prepositional verb data
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
268 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
Correlation with the Gold-standard Data
• For binary tests (simplexasneighbour, WordNet,
ANLT), use the Mann-Whitney U test (rank sum test)
• For other methods, map each output value onto a rank
and apply the Spearman Rank Correlation test (rank
test)
• In each case, calculate the Z score and the probability
of the null hypothesis (i.e. no correlation between the
output of the method and the gold-standard data)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
269 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
Results
• same particle − simplex best performer of similarity-
based methods
• MI best performer of statistical tests
• Question of how to apply the results to the proposed
task of lexical acquisition?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
270 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
Overall Reflections
• Promising results observed for detecting compositionality/decomposability,
but less so for determining the semantic contribution of
individual words in an overall MWE
• What about MWEs where the simplex words don’t occur
with that same POS (e.g. chicken out)
• Effects of polysemy (e.g. run down, run over)
• How to move on to actually semantically decompose an
MWE?
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
271 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
Summary
• How do decomposability and compositionality differ?
• What methods have been proposed for generating gold-
standard compositionality data?
•
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
272 Detecting a Continuum of Compositionality in Phrasal Verbs (McCarthy et al. 2003)
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
273 WRAP-UP
WRAP-UP
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
274 WRAP-UP
Overall Conclusions
• MWEs are frequent, fun and funky in all sorts of ways
• There’s much, much more to MWEs than our old friend
kick the bucket
• More work needs to be done in developing gold-standard
resources to encourage others to enter the fray
• Most of the research problems are far from resolved:
lots of room for everyone to play in!
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
275 WRAP-UP
MWE Resources
• MWE project page: mwe.stanford.edu
• On-line MWE data: mwe.stanford.edu/resources
• On-line bibliographies:
? mwe.stanford.edu/bib.html
? www.ims.uni-stuttgart.de/euralex/bibweb/
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
276 WRAP-UP
References
Aarts, Bas. 1989a. Small Clauses in English: the Non-verbal Types. Berlin: Mouton de Gruyter.
——. 1989b. Verb-preposition constructions and small clauses in English. Journal of Linguistics 25.277–90.
Abeille, Anne. 1988. Light verb constructions and extraction out of NP in a tree adjoining grammar. In Papers ofthe 24th Regional Meeting of the Chicago Linguistics Society .
——. 1990. Lexical and syntactic rules in a tree adjoining grammar. In Proc. of the 28th Annual Meeting of the ACL,292–8.
——. 1995. The flexibility of French idioms: A representation with Lexicalised Tree Adjoining Grammar. In (Everaert
et al. 1995a), chapter 1.
Akimoto, Minoji. 1989. A Study of Verbo-Nominal Structures in English. Tokyo: Shinozaki Shorin.
Aldinger, Nadine. 2004. Towards a dynamic lexicon: Predicting the syntactic argument structure of complex verbs.
In Proc. of the 4th International Conference on Language Resources and Evaluation (LREC 2004), Lisbon,Portugal.
Alexander, Richard J. 1978. Fixed expressions in English: A linguistic, psycholinguistic, sociolinguistic anddidactic study (part 1). Anglistik und Englischunterricht 6.171–88.
—— 1979. Fixed expressions in English: A linguistic, psycholinguistic, sociolinguistic and didactic study (part 2).
Anglistik und Englischunterricht 7.181–202.
Allerton, D.J. 1984. Three (or four) levels of word cooccurrence restriction. Lingua 63.17–40.
Ananiadou, Sophia. 1994. A methodology for automatic term recognition. In Proc. of the 15th InternationalConference on Computational Linguistics (COLING ’94), 1034–8, Kyoto, Japan.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
277 WRAP-UP
Baddorf, Debra S., and Martha W. Evens. 1998. Finding phrases rather than discovering collocations:Searching corpora for dictionary phrases. In Proc. of the 9th Midwest Artificial Intelligence and Cognitive
Science Conference (MAICS’98), Dayton, USA.
Baldwin, Timothy. (to appear). The deep lexical acquisition of english verb-particle constructions. Computer
Speech and Language, Special Issue on Multiword Expressions .
——, Colin Bannard, Takaaki Tanaka, and Dominic Widdows. 2003a. An empirical model of multiwordexpression decomposability. In Proc. of the ACL-2003 Workshop on Multiword Expressions: Analysis,
Acquisition and Treatment, 89–96, Sapporo, Japan.
——, John Beavers, Leonoor van der Beek, Francis Bond, Dan Flickinger, and Ivan A. Sag. 2003b.
In search of a systematic treatment of determinerless PPs. In Proc. of the ACL-SIGSEM Workshopon the Linguistic Dimensions of Prepositions and their Use in Computational Linguistics Formalisms and
Applications, Toulouse, France.
——, Emily M. Bender, Dan Flickinger, Ara Kim, and Stephan Oepen. 2004. Road-testing the EnglishResource Grammar over the British National Corpus. In Proc. of the 4th International Conference on
Language Resources and Evaluation (LREC 2004), Lisbon, Portugal.
——, and Francis Bond. 2002. Multiword expressions: Some problems for Japanese NLP. In Proc. of the 8th
Annual Meeting of the Association for Natural Language Processing (Japan), 379–82, Keihanna, Japan.
——, and Aline Villavicencio. 2002. Extracting the unextractable: A case study on verb-particles. In Proc. of the6th Conference on Natural Language Learning (CoNLL-2002), 98–104, Taipei, Taiwan.
Bame, Ken, 1999. Aspectual and resultative verb-particle constructions with up. Handout for talk presented at theOhio State University Linguistics Graduate Student Colloquium.
Bannard, Colin, 2002. Statistical techniques for automatically inferring the semantics of verb-particle constructions.LinGO Working Paper No. 2002-06.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
278 WRAP-UP
——, Timothy Baldwin, and Alex Lascarides. 2003. A statistical approach to the semantics of verb-particles.In Proc. of the ACL-2003 Workshop on Multiword Expressions: Analysis, Acquisition and Treatment, 65–72,
Sapporo, Japan.
Barkema, Henk. 1996a. The effect of inherent and contextual features on the grammatical flexibility of idioms. In
Synchronic Corpus Linguistics, 69–83. Amsterdam, Netherlands: Rodopi.
——. 1996b. Idiomaticity and terminology: A multi-dimensional descriptive model. Studia Linguistica 50.125–60.
Barker, Ken. 1998. A trainable bracketer for noun modifiers. In Proc. of the 12th Canadian Conference on ArtificialIntelligence, 196–210, Vancouver, Canada.
——, and Stan Szpakowicz. 1998. Semi-automatic recognition of noun modifier relationships. In Proc.of the 36th Annual Meeting of the ACL and 17th International Conference on Computational Linguistics(COLING/ACL-98), 96–102, Montreal, Canada.
Basili, Roberto, Maria Teresa Pazienza, and Paola Velardi. 1993. Semi-automatic extraction of linguisticinformation for syntactic disambiguation. Applied Artificial Intelligence 7.339–64.
Bauer, Laurie. 1979. On the need for pragmatics in the study of nominal compounding. Journal of Pragmatics
3.45–50.
——. 1983. English Word-formation. Cambridge, UK: Cambridge University Press.
Bennis, Hans J., Marcel den Dikken, Peter Jordens, Susan Powers, and Jurgen Weissenborn. 1995.Picking up particles. In Proc. of the 19th Annual Boston University Conference on Language Development
(BUCLD 19), 70–82, Somerville, USA. Cascadilla Press.
Blaheta, Don, and Mark Johnson. 2001. Unsupervised learning of multi-word verbs. In Proc. of the ACL/EACL
2001 Workshop on the Computational Extraction, Analysis and Exploitation of Collocations, 54–60, Toulouse,France.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
279 WRAP-UP
Bolinger, Dwight. 1967. Adjectives in English: Attribution and predication. Lingua 18.1–34.
——. 1971. The Phrasal Verb in English. Harvard, USA: Harvard University Press.
—— (ed.) 1972. Degree Words. the Hague, Netherlands: Mouton.
Bond, Francis, and Satoshi Shirai, 1997. Practical and Efficient Organization of a Large Valency Dictionary.
Handout at the Workshop on Multilingual Information Processing, held in conjunction with NLPRS’97.
Borthen, Kaja, 2003. Norwegian Bare Singulars. Norwegian University of Science and Technology dissertation.
Boucher, Paul, Frederic Danna, and Pascale Sebillot. 1993. Compounds: An intelligent tutoring system
for learning to use compounds in English. Computer Assisted Language Learning (CALL) 6.249–72.
Bouillon, Pierrette. 1996. Mental state adjectives: the perspective of generative lexicon. In Proc. of the 16thInternational Conference on Computational Linguistics (COLING ’96), 143–8, Copenhagen, Denmark.
——, Katharina Boesefeldt, and Graham Russell. 1994. Compound nouns in a unification-based MT system.
In Proc. of the 4th Conference on Applied Natural Language Processing (ANLP), Stuttgart, Germany.
——, and Evelyn Viegas. 1999. The description of adjectives for natural language processing: Theoretical andapplied perspectives. In Proc. of the TALN’99 workshop on Description des adjectifs pour les traitements
informatiques, Cargese, France.
Bourigault, Didier. 1992. Surface grammatical analysis for the extraction of terminological noun phrases. In Proc.of the 14th International Conference on Computational Linguistics (COLING ’92), 977–81, Nantes, Frances.
——. 1993. An endogenous corpus-based method for structural noun phrase disambiguation. In Proc. of the 6thConference of the European Chapter of the Association for Computational Linguistics (EACL’93), Utrecht,Netherlands.
——, Christian Jacquemin, and Marie-Claude LHomme (eds.) London, UK. Recent advances in ComputationalTerminology . Johns Benjamins.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
280 WRAP-UP
Breidt, Elisabeth. 1993. Extraction of V-N-collocations from text corpora: A feasibility study for German. In Proc.of the Workshop on Very Large Corpora: Academic and Industrial Perspectives, Columbus, USA.
Brinton, Laurel. 1985. Verb particles in English: Aspect or aktionsart. Studia Linguistica 39.157–68.
Brugman, Claudia, 1981. The story of over. Master’s thesis, UCLA, Berkeley.
——, 1988. The Syntax and Semantics of HAVE and its Complements. UCLA, Berkeley dissertation.
Busa, Frederica, and Michael Johnston. 1996. Cross-linguistic semantics for complex nominals in thegenerative lexicon. In AISB Workshop on Multilinguality in the Lexicon, Sussex, UK.
Butt, Miriam. 1995. The structure of complex predicates in Urdu. Stanford, USA: CSLI Publications.
Cacciari, Cristina, and Patrizia Tabossi (eds.) 1993a. Idioms: Processing, Structure and Interpretation.Hillsdale, NJ: Lawrence Erlbaum Associates.
——, and ——. 1993b. Preface. In (Cacciari and Tabossi 1993a).
Calzolari, Nicoletta, Charles Fillmore, Ralph Grishman, Nancy Ide, Alessandro Lenci,
Catherine MacLeod, and Antonio Zampolli. 2002. Towards best practice for multiword expressions incomputational lexicons. In Proc. of the 3rd International Conference on Language Resources and Evaluation
(LREC 2002), 1934–40, Las Palmas, Canary Islands.
Cao, Yunbo, and Hang Li. 2002. Base noun phrase translation using Web data and the EM algorithm. In Proc. ofthe 19th International Conference on Computational Linguistics (COLING 2002), Taipei, Taiwan.
Chen, A., K. Kishida, H. Jiang, and Q. Liang. 1999. Automatic construction of a Japanese–English lexicon andits application in cross-language information retrieval. In ACM DL/ACM SIGIR Workshop on MultilingualInformation Discovery and Access (MIDAS).
Chen, K-H., and H-H. Chen. 1994. Extracting noun phrases from large-scale texts: A hybrid approach and itsautomatic evaluation. In Proc. of the 32nd Annual Meeting of the ACL, 234–41.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
281 WRAP-UP
Chen, P. 1986. Discourse and particle movement in English. Studies in Language 10.79–95.
Choueka, Y., S.T. Klein, and E. Neuwitz. 1983. Automatic retrieval of frequent idiomatic and collocational
expressions in a large corpus. Journal of the Association for Literary and Linguistic Computing 4.34–8.
Colombo, L. 1993. The comprehension of ambiguous idioms in context. In (Cacciari and Tabossi 1993a), chapter 8.
Copestake, Ann. 2001. The semi-generative lexicon: limits on lexical productivity. In Proceedings of the 1stInternational Workshop on Generative Approaches to the Lexicon. Geneva.
——. 2002. Implementing Typed Feature Structure Grammars. Stanford, USA: CSLI Publications.
——. 2003. Compounds revisited. In Proc. of the 2nd International Workshop on Generative Approaches to theLexicon, Geneva, Switzerland.
——, and Dan Flickinger. 2000. An open-source grammar development environment and broad-coverage English
grammar using HPSG. In Proc. of the 2nd International Conference on Language Resources and Evaluation(LREC 2000), Athens, Greece.
——, Fabre Lambeau, Aline Villavicencio, Francis Bond, Timothy Baldwin, Ivan A. Sag, and Dan
Flickinger. 2002. Multiword expressions: Linguistic precision and reusability. In Proc. of the 3rdInternational Conference on Language Resources and Evaluation (LREC 2002), 1941–7, Las Palmas, Canary
Islands.
——, Fabre Lambeau, Benjamin Waldron, Francis Bond, Dan Flickinger, and Stephan Oepen. 2004.A lexicon module for a grammar development environment. In Proc. of the 4th International Conference on
Language Resources and Evaluation (LREC 2004), 1111–4, Lisbon, Portugal.
——, and Alex Lascarides. 1997. Integrating symbolic and statistical representations: The lexicon pragmatics
interface. In Proc. of the 35th Annual Meeting of the ACL and 8th Conference of the EACL (ACL-EACL’97),136–43, Madrid, Spain.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
282 WRAP-UP
Cowie, A.P. (ed.) 1998. Phraseology : theory, analysis, and applications. OUP.
——, and P. Howarth. 1996. Phraseology—a select bibliography. International Journal of Lexicography 9.38–51.
Cruse, D. Alan. 1986. Lexical Semantics. Cambridge, UK: Cambridge University Press.
Dagan, I., and K. Church. 1994. Termight: Identifying and translation technical terminology. In Proc. of the 4thConference on Applied Natural Language Processing (ANLP), 34–40, Stuttgart, Germany.
Dagut, M., and B. Laufer. 1985. Avoidance of phrasal verbs: A case for contrastive analysis. Studies in SecondLanguage Acquisition 7.73–9.
Dasgupta, P. 1993. Idiomaticity and Esperanto texts: An empirical study. Linguistics 31.367–86.
Dehe, Nicole. 2000. On particle verbs in English: More evidence from information structure. In Proc. of the 28thWestern Conference on Linguistics (WECOL 1999), 92–105.
——. 2001a. Intonation patterns of particle verb constructions in English. In Proc. of the 31st Conference of the NorthEastern Linguistics Society (NELS 31), 183–97.
——. 2001b. Transitive particle verbs in English: The neutral order. evidence from speech production. In (Dehe andWanner 2001), 165–89.
——. 2002. Particle Verbs in English: Syntax, Information, Structure and Intonation. Amsterdam,
Netherlands/Philadelphia USA: John Benjamins.
——, Ray Jackendoff, Andrew McIntyre, and Silke Urban (eds.) 2002. Verb-particle explorations.
Berlin/New York: Mouton de Gruyter.
——, and Anja Wanner (eds.) 2001. Structural Aspects of Semantically Complex Verbs. Peter Lang.
Den Dikken, M. 1995. Particles: On the Syntax of Verb-particle, Triadic, and Causative Constructions. OxfordUniversity Press.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
283 WRAP-UP
Dias, G., S. Guillore, J-C. Bassano, and J.G. Pereira Lopes. 2000. Combining linguistics with statisticsfor multiword term extraction: A fruitful association? In Proc. of Recherche d’Informations Assistee par
Ordinateur 2000 (RIAO’2000).
Dirven, Rene. 2001. The metaphoric in recent cognitive approaches to English phrasal verbs. metaphorik.de 1.39–54.
Dixon, Robert. 1982. The grammar of English phrasal verbs. Australian Journal of Linguistics 2.149–247.
Dorr, Bonnie J. 1997. Large-scale dictionary construction for foreign language tutoring and interlingual machinetranslation. Machine Translation 12.271–322.
Downing, Pamela. 1977. On the creation and use of English compound nouns. Language 53.810–42.
Erbach, G., and B. Krenn. 1994. Idioms and support-verb constructions in HPSG. In German in Head-driven
Phrase Structure Grammar , ed. by J. Nerbonne, K. Netter, and C. Pollard. Stanford, USA: CSLI LectureNotes.
Everaert, Martin, Erik-Jan van der Linden, Andre Schenk, and Rob Schreuder (eds.) 1995a. Idioms:Structural and Psychological Perspectives. Lawrence Erlbaum Associates.
——, Erik-Jan van der Linden, Andre Schenk, and Rob Schreuder. 1995b. Introduction. In (Everaert et al.
1995a).
Evert, Stefan, and Brigitte Krenn. 2001. Methods for the qualitative evaluation of lexical association measures.In Proc. of the 39th Annual Meeting of the ACL and 10th Conference of the EACL (ACL-EACL 2001),
Toulouse, France.
Fellbaum, Christiane. 1993. The determiner in English idioms. In (Cacciari and Tabossi 1993a), chapter 12.
——. 1998a. Towards a representation of idioms in wordnet. In Proc. of the COLING-ACL’98 Workshop on the Usage
of WordNet in Natural Language Processing Systems, 52–7, Montreal, Canada.
—— (ed.) 1998b. WordNet: An Electronic Lexical Database. Cambridge, USA: MIT Press.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
284 WRAP-UP
Fernando, C., and R. Flavell. 1978. On Idiom: Critical Views and Perspectives. Exeter: University of Exeter.
Ferris, C. 1993. The Meaning of Syntax: A Study of Adjectives of English. London: Longman.
Fillmore, Charles J., and B.T. Sue Atkins. 1994. Starting where the dictionaries stop: The challenge of corpus
lexicography. In Computational Approaches to the Lexicon, ed. by B.T. Sue Atkins and Antonio Zampolli,chapter 13, 349–93. Oxford, UK: Oxford University Press.
——, and Paul Kay. to appear. Construction Grammar . Stanford, USA: CSLI Publications.
——, ——, and Mary C. O’Connor. 1988. Regularity and idiomaticity in grammatical constructions. Language64.501–38.
Finin, T.W. 1980. The semantic interpretation of nominal compounds. In Proc. of the 1st Conference on ArtificialIntelligence (AAAI-80).
Flores d’Arcais, G.B. 1993. The comprehension and semantic interpretation of idioms. In (Cacciari and Tabossi
1993a), chapter 4.
Fraser, B. 1970a. Idioms within a transformational grammar. Foundations of Language 6.22–42.
—— 1970b. Some remarks on the action nominalization in English. In Readings in English Transformational Grammar ,ed. by R.A. Jacobs and P.S. Rosenbaum. Ginn and Company.
—— 1976. The Verb-Particle Combination in English. The Hague: Mouton.
Fung, P., M-Y. Kan, and Y. Horita. 1996. Extracting Japanese domain and technical terms is relatively easy. InProc. of the 2nd International Conference on New Methods in Natural Language Processing , 148–59.
——, and L.Y. Yee. 1998. An IR approach for translating new words from nonparallel, comparable texts. In Proc.
of the 36th Annual Meeting of the ACL and 17th International Conference on Computational Linguistics:COLING/ACL-98 , 414–20.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
285 WRAP-UP
Fung, Pascale. 1995. A pattern matching method for finding noun and proper noun translations from noisy parallelcorpora. In Proc. of the 33rd Annual Meeting of the ACL, 236–43, Cambridge, USA.
——. 1998. Extracting key terms from Chinese and Japanese texts. The International Journal on Computer Processingof Oriental Language, Special Issue on Information Retrieval on Oriental Languages 99–121.
——, and Kathleen McKeown. 1997. Finding terminology translations from non-parallel corpora. In Proc. of the
5th Annual Workshop on Very Large Corpora, 192–202, Hong Kong.
Geeraerts, Dirk. 1995. Specialization and reinterpretation in idioms. In (Everaert et al. 1995a), chapter 3.
Gerber, Laurie, and Jin Yang. 1997. Systran MT dictionary development. In Proc. of the Fifth MachineTranslation Summit (MT Summit V), San Diego, USA.
Gibbs, Raymond W., and Nandini P. Nayak. 1989. Psycholinguistic studies on the syntactic behaviour of idioms.
Cognitive Psychology 21.100–38.
Godby, J., 1994. Two techniques for the identification of phrases in full text.
http://www.oclc.org/research/publications/arr/1994/part1/twotech.htm.
Grefenstette, Gregory. 1999. The World Wide Web as a resource for example-based machine translation tasks.In Translating and the Computer 21: ASLIB’99 , London, UK.
Gries, Stefan T. 1999. Particle movement: A cognitive and functional approach. Cognitive Linguistics 10.105–45.
——, 2000. Towards multifactorial analyses of syntactic variation: The case of particle placement. University of
Hamburg dissertation.
Grishman, Ralph, Catherine Macleod, and Adam Myers, 1998. COMLEX Syntax Reference Manual . ProteusProject, NYU. (http://nlp.cs.nyu.edu/comlex/refman.ps).
Grover, Claire, John Carroll, and Edward Briscoe. 1993. The Alvey Natural Language Tools grammar(4th release). Technical Report 284, Computer Laboratory, Cambridge University, UK.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
286 WRAP-UP
Gueron, J. 1986. Clause union and the verb-particle construction in English. In Proc. of the 16th Conference of theNorth Eastern Linguistics Society (NELS 16).
Haruno, M., S. Ikehara, and T. Yamazaki. 1996. Learning bilingual collocations by word-level sorting. In Proc.of the 16th International Conference on Computational Linguistics (COLING ’96), Copenhagen, Denmark.
Hashimoto, Chikara. 2003. Japanese HPSG: Treatment of syntax and semantics of syntactically complex verbs.In Information Processing Society of Japan SIG Notes. (in Japanese).
Haspelmath, Martin. 1997. From Space to Time in The World’s Languages. Munich, Germany: Lincom Europa.
——. 2002. Understanding Morphology . Arnold Publishers.
Hektoen, E. 1997. Probabilistic parse selection based on semantic cooccurrences. In 5th International workshop on
parsing technologies (IWPT-97), 113–122.
Himeno, Masako. 2001. The nature of compound verbs. Nihongogaku 240.6–15.
Himmelmann, Nikolaus P. 1998. Regularity in irregularity: Article use in adpositional phrases. Linguistic Typology2.315–353.
Hoshi, H., 1994. Passive, Causive, and Light Verbs: A Study of Theta Role Assignment. University of Connecticut
dissertation.
Howarth, P.A. 1996. Phraseology in English Academic Writing . Tbingen: Max Niemeyer.
Huddleston, Rodney, and Geoffrey K. Pullum. 2002. The Cambridge Grammar of the English Language.Cambridge, UK: Cambridge University Press.
Hulstijn, J.H., and E. Marchena. 1989. Avoidance: Grammatical or semantic causes? Studies in SecondLanguage Acquisition 11.241–55.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
287 WRAP-UP
ichiro Kamei, Shin. 1993. Nihongo no yogen-ku soto kan’yo-hyogen no bunrui to sono oyo [A classification ofJapanese verbal idioms]. In Proc. of the 47th National Conference of the Information Processing Society of
Japan, volume 3, 87–8. (In Japanese).
——, Shinko Tamura, and Kazunori Muraki. 1997. Nihongo no yogen-soto kan’yo-hyogen no imi-kukan ni-okerubunpu-zu [A classification of Japanese verbal idioms over a semantic space]. In Proc. of the Third Annual
Meeting of the Japanese Association for Natural Language Processing , 51–4. (In Japanese).
Ikehara, Satoru, Masahiro Miyazaki, Satoshi Shirai, Akio Yokoo, Hiromi Nakaiwa, Kentaro
Ogura, Yoshifumi Ooyama, and Yoshihiko Hayashi. 1997. Nihongo Goi-Taikei – A Japanese Lexicon.
Iwanami Shoten.
IPA, 1987. IPA Lexicon of the Japanese Language for Computers. Tokyo, Japan. (In Japanese).
Isabelle, P. 1984. Another look at nominal compounds. In Proc. of the 10th International Conference onComputational Linguistics (COLING ’84), Stanford, USA.
Ishikawa, K. 2000. A local relation between particles and verbal prefixes in English. English Linguistics 17.249–75.
J. Yoon, K-S. Choi, M. Song. 2001. A corpus-based approach for Korean nominal compound analysis based on
linguistic and statistical information. Natural Language Engineering 7.
Jackendoff, Ray. 1983. Semantics and Cognition. Cambridge, USA: MIT Press.
——. 1995. The boundaries of the lexicon. In (Everaert et al. 1995a), chapter 7.
——. 1996. The proper treatment of measuring out, telicity and perhaps event quantification in English. Natural
Language and Linguistic Theory 14.305–54.
——. 1997a. The Architecture of the Language Faculty . Cambridge, USA: MIT Press.
——. 1997b. Twistin’ the night away. Language 73.534–59.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
288 WRAP-UP
Johansson, C. 1996. Good bigrams. In Proc. of the 16th International Conference on Computational Linguistics(COLING ’96), 592–7, Copenhagen, Denmark.
Johnson, K. 1991. Object positions. Natural Language and Linguistic Theory 9.577–636.
Johnson-Laird, Philip N. 1993. Foreword. In (Cacciari and Tabossi 1993a).
Johnston, Michael, and Frederica Busa. 1996. Qualia structure and the compositional interpretation ofcompounds. In ACL SIGLEX workshop on breadth and depth of semantic lexicons, Santa Cruz, USA.
——, and ——. 1998. The compositional interpretation of nominal compounds. In The Breadth and Depth of
Semantics Lexicons, ed. by Evelyn Viegas. Kluwer Academic.
Jones, Daniel, and Melina Alexa. 1997. Towards automatically aligning German compounds with English word
groups. In New Methods in Language Processing , ed. by Daniel Jones and Harold Somers. London, UK:University College Press.
Justeson, John S., and Slava M. Katz. 1995. Technical terminology: Some linguistic properties and an algorithmfor identification in text. Natural Language Engineering 1.9–27.
Kaalep, Heiki-Jaan, and Kadri Muischnek. 2002. Using the text corpus to create a comprehensive list of
phrasal verbs. In Proc. of the 3rd International Conference on Language Resources and Evaluation (LREC2002), 101–5, Las Palmas, Canary Islands.
Kageura, Kyo. 1997. On intra-term relations of complex terms in the description of term formation patterns. InMelanges de Linguistique Offerts a R. Kocourek , 105–11. Halifax: Les Presses d’ALFA.
——. 1998. A statistical analysis of morphemes in Japanese terminology. In Proc. of the 36th Annual Meeting of the
ACL and 17th International Conference on Computational Linguistics (COLING/ACL-98), 638–45, Montreal,Canada.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
289 WRAP-UP
——, Fuyuki Yoshikane, and Takayuki Nozawa. 2004. Parallel bilingual paraphrase rules for noun compounds:Concepts and rules for exploring Web language resources. In Proc. of the Fourth Workshop on Asian
Language Resources, 54–61, Sanya, China.
Kageyama, Taro. 1993. Grammar and Word Formation. Tokyo, Japan: Hitsuji Shobo.
——. 1999. Word formation. In The Handbook of Japanese Linguistics, 297–325. Blackwell Publishers.
Kaluza, H. 1984. English verbs with prepositions and particles. International Review of Applied Linguistics in
Language Teaching (IRAL) 12.109–13.
Kanzaki, Kyoko. 1997. Lexical semantic relations between adnominal constituents and their head nouns.Mathematical Linguistics 21.53–68. (In Japanese).
——, and Hitoshi Isahara. 1997. Lexical semantics for adnominal constituents in Japanese. In Proc. of the 4thNatural Language Processing Pacific Rim Symposium 1997 (NLPRS’97), 573–6, Phuket, Thailand.
——, Qing Ma, and Hitoshi Isahara. 2000. Similarities and differences among semantic behaviors of japaneseadnominal constituents. In Proc. of ANLP/NAACL2000 Workshop on Syntactic and Semantic Complexity in
Natural Language Processing Systems, 59–68, Seattle, USA.
Kayne, Richard. 1985. Principles of particle constructions. In Grammatical Representation, ed. by J. Gueron, H-G.Obenauer, and J-Y. Pollock, 101–40. Dordrecht, Netherlands: Foris.
Kita, K., Y. Kato, T. Omoto, and Y. Yano. 1994. A comparative study of automatic extraction of collocationsfrom corpora: Mutual information vs. cost criteria. Journal of Natural Language Processing 1.21–33.
——, and H. Ogata. 1997. Collocations in language learning: Corpus-based automatic compilation of collocations
and bilingual collocation concordancer. Computer Assisted Language Learning: An International Journal10.229–38.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
290 WRAP-UP
Kobayasi, Y., T. Tokunaga, and H. Tanaka. 1994. Analysis of Japanese compound nouns using collocationalinformation. In Proc. of the 15th International Conference on Computational Linguistics (COLING ’94),
865–9, Kyoto, Japan.
Langkilde, Irene, and Kevin Knight. 1998. Generation that exploits corpus-based statistical knowledge. In Proc.
of the 36th Annual Meeting of the ACL and 17th International Conference on Computational Linguistics(COLING/ACL-98), 704–710, Montreal, Canada.
Lapata, M., F. Keller, and S. McDonald. 2001. Evaluating smoothing algorithms against plausibility
judgements. In Proc. of the 39th Annual Meeting of the ACL and 10th Conference of the EACL (ACL-EACL2001), Toulouse, France.
——, S. McDonald, and F. Keller. 1999. Determinants of adjective-noun plausibility. In Proc. of the 9th
Conference of the European Chapter of the Association for Computational Linguistics (EACL’99), 30–6.
Lapata, Maria. 2002. The disambiguation of nominalizations. Computational Linguistics 28.357–88.
Lapata, Mirella, and Alex Lascarides. 2003. Detecting novel compounds: The role of distributional evidence.In Proc. of the 10th Conference of the EACL (EACL 2003), Budapest, Hungary.
Lauer, Mark. 1995a. Corpus statistics meet the noun compound: Some empirical results. In Proc. of the 33rdAnnual Meeting of the ACL.
——, 1995b. Designing Statistical Language Learners: Experiments on Noun Compounds. Macquarie University
dissertation.
——, and Mark Dras. 1994. A probabilistic model of compound nouns. In Proc. of the 7th Joint AustralianConference on Artificial Intelligence.
Laufer, B., and S. Eliasson. 1993. What causes avoidance in L2 learning: L1-L2 difference, L1-L2 similarity, or L2complexity? Studies in Second Language Acquisition 15.35–48.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
291 WRAP-UP
Laufer, Batia. 2000. Avoidance of idioms in a second language: The effect of L1-L2 degree of similarity. StudiaLinguistica 54.186–96.
Lee, Hyun Ah, and Gil Chang Kim. 2002. Translation selection through source word sense disambiguation andtarget word selection. In Proc. of the 19th International Conference on Computational Linguistics (COLING
2002), 530–536, Taipei, Taiwan.
Lehrberger, J. 1982. Automatic translation and the concept of sublanguage. In Sublanguage: Studies of Languagein Restricted Semantic Domains, ed. by R. Kittredge and J. Lehrberger. de Gruyter.
—— 1986. Sublanguage analysis. In Analyzing Language in Restricted Domains: Sublanguage Description andProcessing , ed. by Ralph Grishman and Richard Kittredge, chapter 2, 19–38. Lawrence Erlbaum Associates.
Levi, Judith N. 1978. The Syntax and Semantics of Complex Nominals. New York, USA: Academic Press.
Lewis, David D., and W. Bruce Croft. 1990. Term clustering of syntactic phrases. In Proc. of 13th International
ACM-SIGIR Conference on Research and Development in Information Retrieval (SIGIR’90), 385–404, Brussels,Belgium.
Li, Wei, Xiuhong Zhang, Cheng Niu, Yuankai Jiang, and Rohini K. Srihari. 2003. An expert lexiconapproach to identifying English phrasal verbs. In Proc. of the 41st Annual Meeting of the ACL, 513–20,Sapporo, Japan.
Liberman, Mark, and Richard Sproat. 1992. The stress and structure of modified noun phrases in English. In
Lexical Matters – CSLI Lecture Notes No. 24 , ed. by Ivan A. Sag and A. Szabolcsi. Stanford, USA: CSLIPublications.
Liceras, J.M., and L. Dıaz. 2000. Triggers in L2 acquisition: The case of Spanish N-N compounds. StudiaLinguistica 54.197–211.
Lin, Dekang. 1998a. Extracting collocations from text corpora. In Proc. of the COLING-ACL’98 Workshop onComputational Terminology , Montreal, Canada.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
292 WRAP-UP
——. 1998b. Using collocation statistics in information extraction. In Proc. of the Seventh Message UnderstandingConference (MUC-7).
——. 1999. Automatic identification of non-compositional phrases. In Proc. of the 37th Annual Meeting of the ACL,
317–24, College Park, USA.
Lindner, Susan. 1982. What goes up doesn’t necessarily come down: The ins and outs of opposites. In Papers of
the 18th Regional Meeting of the Chicago Linguistics Society , 305–23.
——. 1983. A Lexico-semantic Analysis of Verb-particle Constructions with Up and Out. Indiana, USA: IndianaUniversity Linguistics Club.
Liontas, John I. 2002. Exploring second language learners’ notions of idiomaticity. System 30.289–313.
Lipka, L. 1972. Semantic Structure and Word-Formation: Verb-Particle Constructions in Contemporary English.
Wilhelm Fink Verlag.
Lohse, Barbara, John A. Hawkins, and Tom Wasow, in preparation. Processsing domains in Englishverb-particle constructions.
Ludeling, A. 2001. Particle verbs and similar constructions in German. Stanford, USA: CSLI Publications.
Makai, A. 1972. Idiom Structure in English. Mouton.
Mandala, Rila, Takenobu Tokunaga, and Hozumi Tanaka. 2000. Query expansion using heterogeneous
thesauri. Information Processing and Management 36.361–78.
Manning, Christopher D., and Hinrich Schutze. 1999. Foundations of Statistical Natural Language Processing .
Cambridge, USA: MIT Press.
Marsh, E. 1984. A computational analysis of complex noun phrases in navy messages. In Proc. of the 10thInternational Conference on Computational Linguistics (COLING ’84), 505–8, Stanford, USA.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
293 WRAP-UP
Matsumoto, Yo. 1996. Complex Predicates in Japanese: A Syntactic and Semantic Study of the Notion ‘Word’.Stanford, USA: CSLI & Kurosio Publishers.
——. 1998. The combinatory possibilities in Japanese V-V lexical compounds. Gengo Kenkyu 114.37–83.
Maynard, Diana, and Sophia Ananiadou. 1999. Identifying contextual information for multi-word termextraction. In 5th International Congress on Terminology and Knowledge Engineering (TKE 99), 212–21.
McCarthy, Diana, Bill Keller, and John Carroll. 2003. Detecting a continuum of compositionality in
phrasal verbs. In Proc. of the ACL-2003 Workshop on Multiword Expressions: Analysis, Acquisition andTreatment, Sapporo, Japan.
McIntyre, Andrew, 2001. Introduction to the verb-particle experience. Ms, Leipzig.
McKeown, Kathleen, and Dragomir Radev. 2000. Collocations. In Handbook of Natural Language Processing ,
ed. by Robert Dale, Hermann Moisl, and Harold Somers, chapter 21. Marcel Dekker.
Melamed, I.D. 1998. Empirical methods for MT lexicon development. In Proc. of AMTA’98: Conference of theAssociation for Machine Translation in the Americas, 18–30.
—— 2000. Models of translational equivalence among words. Computational Linguistics 26.221–49.
Mel’cuk, Igor. 1995. Phrasemes in language and phraseology in linguistics. In (Everaert et al. 1995a), chapter 8.
Mel’cuk, Igor. 1998. Collocations and lexical functions. In Phraseology: Theory, Analysis, and Applications, 23–54.
Oxford: Clarendon Press.
——, and Alain Polguere. 1987. A formal lexicon in meaning-text theory (or how to do lexica with words).
Computational Linguistics 13.261–275.
Merkel, M., and M. Andersson. 2000. Knowledge-lite extraction of multi-word units with language filters andentropy thresholds. In Proc. of Recherche d’Informations Assistee par Ordinateur 2000 (RIAO’2000).
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
294 WRAP-UP
——, B. Nilsson, and L. Ahrenberg. 1994. A phrase-retrieval system based on recurrence. In Proc. of the 2ndAnnual Workshop on Very Large Corpora, 99–108.
Meyer, Ralf. 1993. Compound Comprehension in Isolation and in Context. Tubingen: Max Niemeyer Verlag.
Miller, George A., Richard Beckwith, Christiane Fellbaum, Derek Gross, and Katherine J.
Miller. 1990. Introduction to WordNet: an on-line lexical database. International Journal of Lexicography3.235–44.
Mitchell, T.F. 1971. Linguistic ‘goings on’: Collocations and other lexical matters arising on the syntactic record.Archivum Linguisticum 2.35–69.
Miyagawa, Shigeru. 1989. Light verbs and the ergative hypothesis. Linguistic Inquiry 20.659–68.
Moon, Rosamund (ed.) 1991. COBUILD Dictionary of Phrasal Verbs. Collins.
Moon, Rosamund E. 1998. Fixed Expressions and Idioms in English: A Corpus-based Approach. Oxford: ClarendonPress.
Morgan, Pamela S. 1997. Figuring out Figure out : Metaphor and the semantics of the English verb-particleconstruction. Cognitive Linguistics 8.327–57.
Mortimer, C. 1972. Phrasal verbs in conversation. Longman.
Muller, Stefan. 2001. German particle verbs and the predicate complex. In Grammatical Interfaces in HPSG
(Studies in Constraint-Based Lexicalism), ed. by Ronnie Cann, Claire Grover, and Philip Miller. Stanford,USA: CSLI Publications.
Nagata, Masaaki, Teruka Saito, and Kenji Suzuki. 2001. Using the Web as a bilingual dictionary. In Proc. ofthe ACL/EACL 2001 Workshop on Data-Driven Methods in Machine Translation, 95–102, Toulouse, France.
Neeleman, Ad, 1994. Complex Predicates. Utrecht University dissertation.
Nicolas, Tim. 1995. Semantics of idiom modification. In (Everaert et al. 1995a), chapter 9.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
295 WRAP-UP
Niimi, Kazuaki, Youichi Yamaura, and Tokuko Utsuno. 1987. Compound Verbs. Tokyo, Japan: AratakeShuppan. In Japanese.
Nunberg, Geoffrey, Ivan A. Sag, and Tom Wasow. 1994. Idioms. Language 70.491–538.
Odijk, Jan. 2004. Reusable lexical representations for idioms. In Proc. of the 4th International Conference onLanguage Resources and Evaluation (LREC 2004), 903–6, Lisbon, Portugal.
O’Dowd, Elizabeth M. 1998. Prepositions and Particles in English. Oxford University Press.
Ohmori, Kumiko, and Masanobu Higashida. 1999. Extracting bilingual collocations from non-aligned parallel
corpora. In Proc. of the 8th International Conference on Theoretical and Methodological Issues in MachineTranslation (TMI-99), 88–97, Chester, UK.
Peabody, K.W., 1981. Constraints on the productivity of verb-particle combinations. Master’s thesis, Ohio StateUniversity.
Pearce, Darren. 2001a. Synonymy in collocation extraction. In Proc. of the NAACL 2001 Workshop on WordNetand Other Lexical Resources: Applications, Extensions and Customizations, Pittsburgh, USA.
——. 2001b. Using conceptual similarity for collocation extraction. In Proc. of the 4th UK Special Interest Group forComputational Linguistics (CLUK4).
——. 2002. A comparative evaluation of collocation extraction techniques. In Proc. of the 3rd International Conferenceon Language Resources and Evaluation (LREC 2002), Las Palmas, Canary Islands.
Piao, Scott S.L., Paul Rayson, Dawn Archer, Andrew Wilson, and Tony McEnery. 2003. Extractingmultiword expressions with a semantic tagger. In Proc. of the ACL-2003 Workshop on Multiword Expressions:Analysis, Acquisition and Treatment, 49–56, Sapporo, Japan.
Pilz, Klaus Dieter. 1981. Phraseologie. Stuttgart, Germany: Sammlung Metzler.
Pollard, Carl, and Ivan A. Sag. 1994. Head-driven Phrase Structure Grammar . Chicago, USA: The Universityof Chicago Press.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
296 WRAP-UP
Pulman, Stephen G. 1993. The recognition and interpretation of idioms. In (Cacciari and Tabossi 1993a),chapter 11.
Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, USA: MIT Press.
Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik. 1985. A ComprehensiveGrammar of the English Language. London, UK: Longman.
Rackow, Ulrike, Ido Dagan, and Ulrike Schwall. 1992. Automatic translation of noun compounds. In Proc.
of the 14th International Conference on Computational Linguistics (COLING ’92), 1249–53, Nantes, Frances.
Rapp, Reinhard. 1999. Automatic identification of word translations from unrelated English and German corpora.In Proc. of the 37th Annual Meeting of the ACL, 1–17, College Park, USA.
Renouf, A., and J.M. Sinclair. 1991. Collocational frameworks in English. In English Corpus Linguistics, ed. by
K. Aijmer and B. Altenberg, 128–43. London: Longman.
Resnik, Philip, 1993. Selection and Information: A Class-Based Approach to Lexical Relationships. University ofPennsylvania dissertation.
Riehemann, Susanne, 1997. Idiomatic constructions in HPSG. Presented at the 4th International Conference onHPSG.
——, 2001. A Constructional Approach to Idioms and Word Formation. Stanford, USA dissertation.
Ries, K., F.D. Buø, and A. Waibel. 1996. Class phrase models for language modelling. In Proc. of the 4th
International Conference on Spoken Language Processing (ICSLP’96).
Rosario, Barbara, and Marti Hearst. 2001. Classifying the semantic relations in noun compounds via adomain-specific lexical hierarchy. In Proc. of the 6th Conference on Empirical Methods in Natural Language
Processing (EMNLP 2001), Pittsburgh, USA.
Ross, Haj. 1995. Defective noun phrases. In Papers of the 31st Regional Meeting of the Chicago Linguistics Society ,398–440.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
297 WRAP-UP
Sag, Ivan, Timothy Baldwin, Francis Bond, Ann Copestake, and Dan Flickinger. 2002a. Multiwordexpressions: A pain in the neck for NLP. In Computational Linguistics and Intelligent Text Processing: Third
International Conference: CICLing-2002 , ed. by Alexander Gelbuk, 1–15. Hiedelberg/Berlin: Springer-Verlag.
Sag, Ivan A., Timothy Baldwin, Francis Bond, Ann Copestake, and Dan Flickinger. 2002b. Multiword
expressions: A pain in the neck for NLP. In Proc. of the 3rd International Conference on Intelligent TextProcessing and Computational Linguistics (CICLing-2002), 1–15, Mexico City, Mexico.
——, Thomas Wasow, and Emily Bender. 2003. Syntactic Theory: A Formal Introduction, 2nd Edition. Stanford,
USA: CSLI Publications. Chapter 8: The Structure of the Lexicon.
Salton, Gerald, and Maria Smith. 1990. On the application of syntactic methodologies in automatic textanalysis. In Proc. of 13th International ACM-SIGIR Conference on Research and Development in Information
Retrieval (SIGIR’90), 137–50, Cambridge, USA.
Sawyer, Joan H., 1999. Verb Adverb and Verb Particle Constructions: Their Syntax and Acquisition. BostonUniversity dissertation.
Schenk, Andre. 1995. The syntactic behavior of idioms. In (Everaert et al. 1995a), chapter 10.
Schenk, Anree. 1986. Idioms in the Rosetta machine translation system. In Proc. of the 11th InternationalConference on Computational Linguistics (COLING ’86), Bonn, Germany.
Schone, Patrick, and Dan Jurafsky. 2001. Is knowledge-free induction of multiword unit dictionary headwords
a solved problem? In Proc. of the 6th Conference on Empirical Methods in Natural Language Processing(EMNLP 2001), 100–108.
Schreuder, Robert. 1990. Lexical processing of verbs with separable particles. Yearbook of Morphology 3.65–79.
Sekine, Satoshi, Jeremy J. Carroll, Sophia Ananiadou, and Jun’ichi Tsujii. 1992. Automatic learningfor semantic collocation. In Proc. of the 3rd Conference on Applied Natural Language Processing (ANLP).
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
298 WRAP-UP
Selkirk, Elisabeth O. 1982. The Syntax of Words. Cambridge, USA: MIT Press.
Selver, P. 1957. English Phraseology, a Dictionary Containing more than 5,000 Idiomatic and Colloquial Words and
Expressions. London : J. Brodie.
Shimamura, Reiko. 1998. Lexicalization of syntactic phrases: The case of genitive compounds like Woman’s
Magazine. In Proc. of the Kanda University of International Studies Graduate School of Language SciencesCentre of Excellence in Linguistics (COE) International Workshop.
——. 2001. The A-N expression within the compound and the phrase/word distinction. In Proc. of the Linguistics
Society of America (LSA) Annual Meeting .
Shimohata, Sayori, Toshiyuki Sugio, and Junji Nagata. 1997. Retrieving collocations by co-occurrences andword order constraints. In Proc. of the 35th Annual Meeting of the ACL and 8th Conference of the EACL
(ACL-EACL’97), 476–81, Madrid, Spain.
Shirai, Satoshi, Yoshifumi Ooyama, Shinobu Takechi, Keiko Wakebe, and Hiroshi Aizawa. 1998.
Compiling Japanese and English corpus for compound verbs of Japanese origin. In Proc. of the 57th AnnualMeeting of IPSJ , 267–8, Nagoya, Japan. In Japanese.
Side, Richard. 1990. Phrasal verbs: sorting them out. ELT Journal 44.144–52.
Sinclair, John M. 1987. Collocation: A progress report. In Language Topics: Essays in Honour of Michael Halliday ,ed. by R. Steele and T. Threadgold, volume II, 319–31. John Benjamins.
Smadja, Frank. 1993. Retrieving collocations from text: Xtract. Computational Linguistics 19.143–77.
——, Kathleen R. McKeown, and Vasileios Hatzivassiloglou. 1996a. Translating collocations for bilinguallexicons: A statistical approach. Computational Linguistics 22.1–38.
——, ——, and ——. 1996b. Translating collocations for bilingual lexicons: A statistical approach. ComputationalLinguistics 22.1–38.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
299 WRAP-UP
Smadja, Frank A., and Kathleen R. McKeown. 1990. Automatically extracting and representing collocationsfor language generation. In Proc. of the 28th Annual Meeting of the ACL, 252–9.
Smith, Jeffrey D. 1999. English number names in hpsg. In Lexical and Constructional Aspects of LinguisticExplanation, ed. by Andreas Kathol, Jean-Pierre Koenig, Gert Webelhuth, and eds., 145–160. Stanford, USA:
CSLI Publications.
Soehn, Jan-Philipp, and Manfred Sailer. 2003. At first blush on tenterhooks. About selectional restrictions
imposed by nonheads. In Proceedings of Formal Grammar 2003 , ed. by Gerhard Jager, Paola Monachesi,Gerald Penn, and Shuly Winter, 149–161.
Sproat, Richard W. 1994. English noun-phrase accent prediction for text-to-speech. Computer Speech and
Language 8.79–94.
——, and Mark Y. Liberman. 1987. Toward treating English nominals correctly. In Proc. of the 25th Annual
Meeting of the ACL, Stanford, USA.
Stock, Oliverio, Jon Slack, and Andrew Ortony. 1993. Building castles in the air: Some conceptual andtheoretical issues in idiom comprehension. In (Cacciari and Tabossi 1993a), chapter 10.
Stock, Oliviero. 1987. Getting idioms into a lexicon based parser’s head. In Proc. of the 25th Annual Meeting ofthe ACL, 52–8, Stanford, USA.
Stowell, Tim, 1981. Origins of Phrase Structure. MIT dissertation.
Stvan, Laurel Smith, 1998. The Semantics and Pragmatics of Bare Singular Noun Phrases. Northwestern Universitydissertation.
Sweetser, Eve. 1990. From etymology to pragmatics. Cambridge, UK: Cambridge University Press.
Tanaka, Takaaki. 2002. Measuring the similarity between compound nouns in different languages using non-parallel
corpora. In Proc. of the 19th International Conference on Computational Linguistics (COLING 2002), 981–7,Taipei, Taiwan.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
300 WRAP-UP
——, and Timothy Baldwin. 2003a. Noun-noun compound machine translation: A feasibility study on shallowprocessing. In Proc. of the ACL-2003 Workshop on Multiword Expressions: Analysis, Acquisition and
Treatment, 17–24, Sapporo, Japan.
——, and ——. 2003b. Translation selection for Japanese-English noun-noun compounds. In Proc. of the NinthMachine Translation Summit (MT Summit IX), 89–96, New Orleans, USA.
——, and Yoshihiro Matsuo. 1999. Extraction of translation equivalents from non-parallel corpora. In Proc. ofthe 8th International Conference on Theoretical and Methodological Issues in Machine Translation (TMI-99),
109–19, Chester, UK.
ter Stal, W., and P. van der Vet. 1994. Two-level semantic analysis of compounds: A case study in linguisticengineering. In Papers from the 4th CLIN Meeting .
Toutanova, Kristina, and Christoper D. Manning. 2000. Enriching the knowledge sources used in a maximumentropy part-of-speech tagger. In Proc. of the Joint SIGDAT Conference on Empirical Methods in Natural
Language Processing and Very Large Corpora (EMNLP/VLC-2000), Hong Kong, China.
Tschichold, Cornelia. 1995. English multi-word lexemes in a lexical database. In Proc. of the ESSLI Workshopon the Computational Lexicon, ed. by M. Felisa Verdejo.
——, 1998. Multi-word Units in Natural Language Processing . University of Basel dissertation.
Tsuji, Keita, and Kyo Kageura. 2001. Extracting morpheme pairs from bilingual terminological corpora.Terminology 7.101–14.
Cermak, F. 1988. On the substance of idioms. Folia Linguistica 22.413–38.
—— 1994. Idiomatics. In Prague School of Structural and Functional Linguistics, ed. by P.A. Luelsdorff, 185–95.Amsterdam and Philadelphia: John Benjamins.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
301 WRAP-UP
Uchiyama, Kiyoko, and Timothy Baldwin. 2004. A machine learning method for disambiguating Japaneseverb compounds. In Proc. of the 10th Annual Meeting of the Association for Natural Language Processing
(Japan), Tokyo, Japan. (in Japanese).
——, ——, and Shun Ishizaki. to appear. Disambiguating japanese compound verbs. Computer Speech andLanguage, Special Issue on Multiword Expressions .
Van de Voort, Marlies E.C., and Wietske Vonk. 1995. You don’t die immediately when you kick an emptybucket: A processing view on semantic and syntactic characteristics of idioms. In (Everaert et al. 1995a),
chapter 12.
van der Linden, Erik-Jan. 1992. Incremental processing and the hierarchical lexicon. Computational Linguistics18.219–38.
van Gestel, Frank. 1995. En bloc insertion. In (Everaert et al. 1995a), chapter 4.
Van Lancker, Diana, and Gerald J. Carter. 1981. Idiomatic versus literal interpretations of ditropicallyambiguous sentences. Journal of Speech and Hearing Research 24.64–9.
——, and Dale Terbeek. 1981. Disambiguation of ditropic sentences: Acoustic and phonetic clues. Journal ofSpeech and Hearing Research 24.330–5.
Vanderwende, Lucy. 1994. Algorithm for automatic interpretation of noun sequences. In Proc. of the 15th
International Conference on Computational Linguistics (COLING ’94), Kyoto, Japan.
Velardi, Paola, Maria Teresa Pazienza, and Michela Fasolo. 1991. How to encode semantic knowledge:
A method for meaning representation and computer-aided acquisition. Computational Linguistics 17.153–70.
Villavicencio, Aline. 2003a. Verb-particle constructions and lexical resources. In Proc. of the ACL-2003 Workshopon Multiword Expressions: Analysis, Acquisition and Treatment, 57–64, Sapporo, Japan.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
302 WRAP-UP
——. 2003b. Verb-particle constructions in the world wide web. In Proc. of the ACL-SIGSEM Workshop on theLinguistic Dimensions of Prepositions and their Use in Computational Linguistics Formalisms and Applications,
Toulouse, France.
——, and Ann Copestake, 2002a. Dictionaries and the regularities in phrasal verbs. LinGO Working Paper No.
2002-03.
——, and Ann Copestake, 2002b. The nature of idioms. LinGO Working Paper No. 2002-04.
——, and Ann Copestake, 2002c. Phrasal verbs and the LinGO-ERG. LinGO Working Paper No. 2002-01.
——, and Ann Copestake, 2002d. The treatment of multiword expressions in the LKB system. LinGO WorkingPaper No. 2002-05.
——, and Ann Copestake. 2002e. Verb-particle constructions in a computational grammar of English. In Proc.
of the 9th International Conference on Head-Driven Phrase Structure Grammar (HPSG-2002), Seoul, SouthKorea.
Wacholder, Nina, and Peng Song. 2003. Toward a task-based gold standard for evaluation of np chunks andtechnical terms. In Proc. of the 3rd International Conference on Human Language Technology Research and4th Annual Meeting of the NAACL (HLT-NAACL 2003), 189–96, Edmonton, Canada.
Wasow, Thomas. 2002. Postverbal Behavior . Stanford, USA: CSLI Publications.
Wehrli, Eric. 1998. Translating idioms. In Proc. of the 36th Annual Meeting of the ACL and 17th InternationalConference on Computational Linguistics: COLING/ACL-98 , 1388–92, Montreal, Canada.
Weinreich, Uriel. 1963. Lexicology. In Current Trends in Linguistics, ed. by T. Sebeok, volume I, 60–93. TheHague: Mouton.
Widdows, Dominic, and Beate Dorow. 2002. A graph model for unsupervised lexical acquisition. In Proc. of the19th International Conference on Computational Linguistics (COLING 2002), 1093–9, Taipei, Taiwan.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf
303 WRAP-UP
Woods, William. 1972. The Lunar sciences natural language information system. Final Report, Bolt, Beranek andNewman, Cambridge, MA.
Wu, Dekai. 1993. Approximating maximum-entropy ratings for evidential parsing and semantic interpretation. In
Proc. of the 13th International Joint Conference on Artificial Intelligence (IJCAI-93), 1290–6.
Wunderlich, Dieter. 1983. On the compositionality of German prefix verbs. In Meaning, Use and the Interpretation
of Language, ed. by R. Bauerle, C. Schwarze, and A. von Stechow, 452–65. De Gruyter.
Wurmbrand, Susi, 2000. The structure(s) of particle verbs. Talk given at the DGfS 2000 (AG3: Semantisch
komplexe Verben und ihre Argumentstruktur), Marburg, Germany.
Y, G., and M. Johnson. 1980. Metaphors we live by . University of Chicago Press, Chicago.
Yamamoto, Mikio, and Kenneth W. Church. 2001. Using suffix arrays to compute term frequency anddocument frequency for all substrings in a corpus. Computational Linguistics 27.1–30.
Zeller, Jochen, 1997. Particle verbs and a theory of late lexical insertion. University of Frankfurt.
——, 1999. Particle verbs, local domains, and a theory of lexical licensing . University of Frankfurt dissertation.
——. 2001. Particle Verbs and Local Domains. Amsterdam: John Benjamins.
Zhang, Lei, Jianfeng Gao, and Ming Zhou. 2000. Extraction of Chinese compound words – an experimental
study on a very large corpus. In Proc. of the 2nd Chinese Language Processing Workshop, ACL 2000 .
Zwicky, Arnold. 1985. Clitics and particles. Language 61.283–305.
www.csse.unimelb.edu.au/~tim/pubs/altss2004.pdf